Image-based Motion-driven Facial Texture

Size: px
Start display at page:

Download "Image-based Motion-driven Facial Texture"

Transcription

1 Image-based Motion-driven Facial Texture Bing Zhang and Hai Tao and Alex Pang Computer Science and Computer Engineering University of California, Santa Cruz 1 INTRODUCTION Facial animation is a fundamental and difficult topic in computer graphics. Its applications include character animation, computer games, video conferencing, video telephony, and avatars etc. The difficulty of facial animation comes mainly from the following factors: complex geometric face models; countless creases and wrinkles; delicate nuances in color and texture; complicated facial movements; human s instinctive detection of subtle variations in facial expressions. Approaches in facial animation fall into two major categories, geometric manipulation based and image manipulation based. Geometric manipulation includes keyframing and geometric interpolation, parameterization, physics-based muscle modeling, pseudo or simulated muscle modeling etc. Image manipulation includes image morphing, texture manipulation, image blending and vascular expressions etc. Because of the complicated facial structure and movements, geometric manipulation is hard to control to get realistic animation. Image manipulation could be combined to improve the animation effect.

2 1.1 Related work Volker Blanz and Thomas Vetter [1] introduced a technique for modeling textured 3D faces. Starting from a set of 3D face models, they derive a morphable face model by transforming the shape and texture of the models into a vector space representation. New faces and expressions can be modeled by forming linear combinations of the prototypes. 3D faces can either be generated from one or more photographs, or modeled directly through the user interface. Shape and texture can be modified in a natural way. Frederic Pighin et. al. [6] presents a technique for creating photorealistic textured 3D facial models from photographs of a human subject, and creating smooth transitions between different facial expressions by morphing between these different models. Starting from several uncalibrated views, they deform the generic 3D face model with recovered 3D information, then extract texture from images for natural looking animations. Instead of creating new facial animations from scratch for each of new models, Junyong Noh and Ulrich Neumann [7] use existing animation data in the form of vertex motion vectors, and transfer them to target models with different geometric proportions and mesh structures. Facial expressoins exhibit not only facial feature motions, but also changes in illumination and appearance caused by the surface normal. Zicheng liu et. al. [4] proposed a skin-color independent (ERI to capture the illumination change of one person s expression and map it to other persons. 1.2 IBMDT We propose image-based motion driven textures (IBMDT as shown in Fig.1. We are not only interested in recover realistic facial animation, but also the explicit descrip-

3 tors about motion vertices and motion textures, so that facial animation could be generated on the fly. IBMDT avoids complicated mechanism behind physical face structures. Using only single view video sequences, IBMDT recovers rigid and non-rigid motions, and finds expression function from motion vertices to motion textures, which are motion driven textures. The project is under construction. The rest of this report is organized as follows: Section II describes preprocessing; Section III talks about registration; Section IV discusses motion capture processes. Section V presents motion driven textures. Section IV is the conclusion about the future work. Fig. 1. IBMDT structure 2 PREPROCESSING 2.1 Data Collection We obtained a set of laser scan 3D face models from Experimental Psychology lab. at UCSC. Focusing on one of Trevor Chen s neutral face model, we captured his facial expressions with three camcorders at different angles. Note the 3D models and video sequences are taken about three months apart. Starting with a neutral expression, each section of video sequences is of different facial expressions. The camcorders were not

4 mechanically synchronized, so we used a laser pointer to help align each segment of video sequences around the same time, though currently only sequences from the front view are used for experiments. Markers were sticked on the face around wrinkled areas so that we could obtaion both vertex motions and texture motions from single-view image sequences. 2.2 Mesh Simplification The mesh of the 3D face model is very dense. It has 191,594 vertices and 383,184 polygons. In order to speed up the rendering, we used a mesh similification software, Michael Garland s QSlim 2.0 [2]. It is very efficient, but a little problem is that feature points on the mesh can not be selectively preserved. In addition, we also programed to cut unnecessary polygons by thresholds, and delete unnecessary ones by interactive vertex picking. Extra work was done for conversions of data formats. After simplification, for example, the mesh size is reduced to 4,811 vertices and 9203 polygons. 2.3 Graphical User Interface We built a user friendly interface, Fig. 2. There are two windows, one for 3D face model, or graphics window (GW on the left, and one for image, or image window (IMG on the right. Mapping can be set up between 3D vertices in GW and 2D pixels on IMG. We can also pick up vertices on GW or mark pixels on IMG to export. Other major functions in GUI include editing for mapping list, synchronized 3D-2D matching, texture mapping, jump from frame to frame or animation etc. 3 REGISTRATION At this stage, we try to align the 3D mesh with captured 2D images. Starting from neutral face, we match 3D feature vertices in graphics window and corresponding 2D

5 Graphics window Image window Fig. 2. Graphical User Interface. Graphics window vs image window feature pixels in image window by interactively picking and marking (see Fig. 2. These vertex and pixel feature points form a 3D-2D pair set, from which we can calibrate camera with modified Tsai s algorithm and recover camera s intrinsic and extrinsic parameters. We further optimize camera s rotation matrix with descent gradient and Newton s method. At last, with calibrated and optimized camera projection model, we reproject all vertices to the image to find corresponding texture coordinates in order to perform texture mapping on 3D model. 3.1 Tsai s camera calibration Assume pxl = K(RWT projection model, where pxl is image pixel coordinate, W is 3D world coordinate, K is camera matrix, R is camera rotation matrix, and T is camera translation matrix. Given matched 3D-2D feature point pairs, we try to recover camera s intrinsic and extrinsic parameters. For camera s intrinsic parameters, the camera distortion is ignored, and the center of the image is the principal point. Then camera matrix K is:.

6 <... Camera s extrinsic parameters are translation matrix T, and rotation matrix R,. Assume that is the 3D world coordinate of a point, and is the corresponding 2D pixel coordinate on the image, we have " $ From Eq.1 and Eq.2, we obtain * & %& ', If we denote the aspect ratio as -, 0 1 ' ( /., and let,,, 32 4-, 35 4-, 36 4-, (1 (2 (3, and '9:9;9 37, then Tsai s camera calibration algorithm can be implemented as the following steps: 1. construct A as: 9;9:9 =>=?=>=?=@=,=A=B=A=B=AC=>=,= 2. solving Av=0, we get the eigenvector corresponding to the smallest eigenvalue of AA, and ED F D HG IG or D JKG G - HG IG 3. obtain 37 0 and D - DL-. G IG. Here,,

7 .. D 4. obtain rotation matrix R. & %& % % D - %& % % 5. to make R orthonormal, we compute R=UDV with singular value decomposition, and update R= R=UIV, where I is 3x3 identity matrix. 6. next, the sign of D is determined by *. 7. given R, Here,, and D, we construct F and b to solve " and, i.e.. =A = ' ' = H= The solution is 9;9:9 = % % 9;9;9 = (= ' ' =. C 3.2 Implementation and Modification With known camera matrix K, rotation matrix R and translation matrix T, we reproject all vertices of 3D face mesh to image plane with equation pxl = K(RWT. By comparing the original pixel coordinates with reprojected ones for the 3D-2D pairs, we realize that there are some problems in Tsai s calibration algorithm. First, Tsai s camera calibration algorithm is very sensitive to the set and accuracy of matched feature point

8 D - - D - - D - - pairs; Secondly, we get unexpected negative texture coordinates from reprojection. For the experimental 3D-2D pair set (Fig. 2, the mean square error is as high as 218 pixels. In the beginning, we tried to normalize 3D world coordinates and 2D pixel coordinates to improve reprojection, but there is no much difference. Then we tried to get camera matrix K from OpenCV camera calibration filter instead of from Tsai s algorithm, because Zhengyou Zhang s calibration pattern [8] was used when we recorded the video sequences. The filter collected 30 frames of the calibration pattern, then out- put,, distortion factors, and principal point. We modified Tsai s algorithm with OpenCV s and. The result is better, and the mean square error of reprojection is 1.56 pixels. But it is still not satisfactory even with later optimization step to improve rotation matrix R. We realize that orthonormalization of rotaion matrix R in Tsai s algorithm (step 5 might be a problem for us to find a fitting projection, therefore we recalculate R without orthonormalization. It turns out to be good for both Tsai s original algorithm and modified version with OpenCV camera parameters. The mean square error of reprojection becomes 0.55 pixels. It proves that step 5 is really a blind step. However, R is NOT orthonormal as required here. More actions should be taken to solve this problem. As we know, rotation matrix R consists of three components, -, and D, which are rotation angles around X, Y, Z axes. Therefore we have D D D D D D D By observing the above matrix, we can easily derive from item. The choice for deriving - and D is a little tricky. Intuitively, matrix R should be close to orthonormal,

9 D D D D D D and items on the main diagonal should best represent its three component angles. Thus we choose to derive - and D from and.. The exception case is when. However such case is rare and it can be easily avoided by adjusting the camera s angle. If we then recalculate R with the derived -, and D based on the above matrix, R won t lose much its own identity and it is surely orthonormal. Therefore, instead of step 5 in Tsai s algorithm, we update rotation matrix R with newly derived -, and D. The reprojection result turns out to be very good. Experiments also show the choice on deriving - and D is best. 3.3 Optimization Optimization is also used to improve rotation matrix R with respect to -, and D. For simplicity, we normalize Eq.(1 and Eq.(2 as: & ' ' ' ( (4 (5 Here, A - D - D ' A - - ' A - ' A - D - D - - -

10 Improvement on R becomes a minimization problem: - D - D (6 - D - D - (7 Here is reprojected normalized pixel coordinate after camera calibration, and corresponds to the original normalized pixel coordinate in the picked 3D-2D pairs. In the beginning of optimization, Descent Gradient is used. Newton s method is also applied when -, and D are close to the solution. The optimization provides one order of magnitude reduction in mean square error. Method of Descent Gradient Suppose the function of is to be minimized, and it has continuous first partial derivatives. The method of descent gradient is defined by the iterative algorithm > (8 where is a non-negative scalar minimizing f(. From the point, we search along the direction of the negative gradient - to a minimum point. Applying descent gradient optimization method to Eq.(7, we have - D - D K@ (9

11 Rotation matrix R could be improved by iteratively updating R s three component angles with D (10 D D Newton s Method With Newton s Method, the function being minimized is approximated locally by a quadratic function, and this approximate function is minimized exactly. Near, can be approximated by the truncated Taylor series: $ (11 The rihgt-hand side is minimized at > $ B $ (12 Eq.12 is the pure form of Newton s method. Assume that at a relative minimum point,, the Hessian matrix $ is positive definite, then if has continuous second partial derevatives, the Newton s method is well defined near the solution. Therefore the Newton s method is used at the second optimization level for rotation matrix R when the Hessian matrix is positive definite. When the Hessian matrix is not positive definite, a scaled portion of identity matrix is added to the Hessian matrix to make it positive. Applying Newton s method to Eq.(7, we have - D (13

12 Rotation matrix R could then be improved by iteratively updating R s three component angles with - D - D - D - D ( Implementation Results Fig. 3. Registration result. Fig.3 shows the final result in the registration stage. The same purple marks on bare 3D model (left and texture-mapped 3D model (right demonstrate that the texture are well aligned onto 3D mesh. Note the original image is shoot from the front, therefore the texture mapping here is not view-dependent. 4 MOTION CAPTURE The facial motion on the image can be classified as rigid motion and non-rigid motion. Rigid motion is caused by the body and/or head movements with respect to the camera,

13 and non-rigid motion is caused by the movements of facial muscles, which also lead to different facial expressions. In order to capture both rigid and non-rigid motions, we put markers on the model s face surrounding the highly wrinkled areas. By tracking those markers and some feature points from frame to frame, we can recover rigid and non-rigid motions. Vertex motions from frame to frame can thuse be derived with gaussian basis functions. The steps at this stage are: 1. Detect markers (optional; 2. Determine feature points for rigid motion either interactively picking from GUI or automatically from marker detection. The feature points should be relatively still, and they may be markers or non-markers; 3. Track feature points for rigid motion frame-by-frame; 4. Determine feature points for non-rigid motion either interactively picking from GUI or automatically by marker detection. The feature points should be relatively active, and they also may be markers or non-markers; 5. Track feature points for non-rigid motion frame-by-frame; 6. Interpolate vertex motions with gaussian basis functions Note that tracking feature points picked or detected may not be well aligned on mesh vertices. In such cases, we will adjust tracking feature points to the closest pixels which are sitted right on 3D vertices. 4.1 Marker Detection The markers used are in red, green and blue. Because the percepted color is better represented by HSV color space than by RGB color space, we first convert the image from RGB to HSV. By sampling some red markers, green markers and blue markers we

14 obtain means and variances in HSV space for each individual color set. We then impose low and high thresholds based on means and variances to detect red, green and blue markers. Experiments show that detecting with only one channel is not enough, so we use all three channels, hue, satuation and value. One problem in implementation is that red markers are harder to detect because of the skin tone. Some dark spots on the face may also be misdetected as red markers. On the other side, green markers are easier to detect than other two kinds of markers. 4.2 Tracking Starting with the neutral frame, the tracking on the feature points are performed frameby-frame. For a feature point on the i th frame, we use a correlation window (2n1 by (2n1 centered at this point. Then we select a search area of size (2m1 by (2m1 around the same location in the next frame. Comparing the correlation window in the i th frame with each correlation window in the following frame within the searching area, we pick one which has the biggest correlation coefficient and mark it as a match to the feature point in the frame. Experiments show that the size of correlation window and the size of searching area are both critical to the tracking quality. The good size for the correlation window should be close to the marker s size. The searching area cannot be too much, because the tracking might be confused with nearby other markers and it is time consuming; on the other side, the searching area cannot be too small, becaused the tracked points might be out of field. When we do tracking for rigid motions, the initial tracking set of feature points are critical. We are in favor of relative areas such as ears, inner eye corners, nose tip etc. However, when we do tracking for non-rigid motions, we are in favor of relative areas which usually surround wrinkles. Selections for tracking sets should all avoid over coplanar of feature points. Fig. 4 is a series of frames showing tracked feature points.

15 Fig. 4. Tracking on feature points 4.3 Rigid Motion Body and/or head movements are classified as rigid motions. Without any loss, we assume that rigid motions were relatively caused by camera motion instead and thus there were no body or head movements. In other words, when only rigid motions are involved, mesh vertices won t move but their projection on image planes might change as the camera moves. In Fig.5 (a, W is 3D vertex on the face mesh, as camera moves from position C1 to position C2, W s projection on the image moves from p to p. (a (b Fig. 5. Rigid motion vs. non-rigid motion We also assume that rigid motions are continuous and smooth from frame to frame. Therefore, taylor expansion is used to derive camera s motions. Because camera motion

16 - D - consists of translation along X/Y/Z axes, and rotation around X/Y/Z axes. following the notations in Eq.(4 and Eq.(5, we have Jacobian matrix (15 From frame k to frame (k1, because the hessian matrix is zero, we have Therefore from Eq.(16, camera s motion from frame k to frame k1 is $ $ D (16 ( Non-rigid Motion At this step, we first track a set of feature points which are around highly active motion areas on the face. As before images change from frame to frame. Besides rigid motions, there are also non-rigid motions due to facial movements such as smiling, frowning etc. In Fig.5 (b, we observe that 3D mesh vertex moves from W to W. Due to dual effects of rigid motion and non-rigid motion, the image projection changes from p to p. Dash lines in the figure are for comparison with the pure rigid motion case in (a. Note in fact we don t know the depth of W,or where is the 3D position of W. The only thing we are sure is that W is located somewhere close to W along the projection line of C2p. Considering that a lot of facial motions are along the face s tangent

17 < surface, we assume that W is on the intersection of C2p and W s tangent plane as shown in Fig.6 (a, where is vertex W s normal. Although people may argue about this method, it is simple. Further, we don t really care if it is accurate at this step. We will see it in the following sections. (a (b Fig. 6. Heuristic solution to non-rigid motion Given W(X0,Y0,Z0 and (Nx,Ny,Nz for the k th frame, normalized pixel coordinate (PX,PY from tracking and camera motion (or rigid motion for the (k1 th frame, we construct A as: % % % ' and b as:

18 ' % % ' K Then the solution to W (X,Y,Z in the (k1 th frame is: < < <. Starting with the neutral face, normals of feature vertices on the first frame are easily accessible. However for the following frames, how do we estimate normals as the vertices move with non-rigid motions? We could repeatedly use normals from the first frame, but it would lead the new vertices move away from the face mesh further and further. One heuristic method is used as shown in Fig.6 (b, which is the enlarged version of (a. In the figure, triangles are supposed to be the neutral face mesh, and the blue dash line extended from C2p W intersects one of triangles at Q. We can obtain Q s normal on the triangle with tri-linear interpolation. By assigning new normal = for W, it will make the new vertices follow the face skeleton to some extend. Certainly there are other options. Constrains may also be imposed on how far away vertices can move. When the limitations are reached, motions along normal may also be considered. 4.5 Vertex Motion Disregarding the complexity of facial muscle structures, vertex motions can be understood as a simulation to the real physical motions from different facial expressions. We use gaussian basis functions to simplify the derivation of vertex motions. The feature

19 points picked for non-rigid motion tracking are used as control points for interpolation on vertex motions of the model mesh. Note that vertex motions do not involve rigid motions. They are caused by non-rigid motions. Fig. 7. Gaussian Basis Function Gaussian Basis Functions As shown in Fig.7, there are n control points marked in blue at their neutral positions %,...,. Their relative motions to the neutral positions in one following frame are,...,. The circles with gaussian curves mean that control points affect the surrounding areas in a way similar to gaussian functions. The red marked point p in the figure is like any other common vertex on the mesh, and its motion is determined by (18 Here / are coefficients for control points. They can be derived from (19 or let

20 < 9;9:9 9;9:9 9;9:9 9:9;9 9;9:9 9;9;9 9;9:9 and 9:9;9 Then the solution to coefficients is 9;9;9 < < <. 5 Motion Driven Textures Different facial expression caused by different and complicated muscle movements, and different textures are driven by different but simple vertex motions. In order to learn the relationship between textures and vertex motions, neuron network is used to do training. In our experiment, we use feedforward neuron networks. The training algorithm uses a gradient descent technique, backpropagation, which involves performing computations backwards through the network [5]. Proper trained backpropagation networks tend to give reasonable answers when presented with inputs that they have never seen. Typically, a new input leads to an output similar to the correct output if input vectors used in training are similar to the newly

21 presented input. This generalization property makes it possible to train a network on a representative set of input/target pairs and get good results without training the netwrok on all possible input/output paris. We know that people s expressions are numerous, but they are highly correlated. Same is true for vertex motions. Therefore, neuron network provides a convenient way for us to extract the driven force behind the vertex motions and textures. The steps in the training process are 1. Assemble the training data. Because both motion vertices and textures are highly correlated, we first apply Principal Component Analysis (PCA, [3] on each of them to reduce redundancy. The input of network is related to the motion vertices, and the output is related to the textures. 2. Create the network object. 3. Train the network. 4. Simulate the network response to new inputs. At this stage, we fulfill the mechanism for motion driven textures. When we feed the network with new motion vertices, we are supposed to get new desired textures. 5.1 PCA on Vertex With GUI, we pick n vertices in an interested area (such as highly wrinkled area from the first neutral mesh. Then we align their 3D world coordinates in a column vector, V1 = [,,,...,,, ]. For the following (m-1 frames, we know the 3D coordinates of the same vertex set from non-rigid motions, and we put them similarly into column vectors V2,...,Vm. Because these m vectors are highly correlated, we apply principal component analysis on them. Then we get some number, i, of significant principal components, which could be far less than m. We can reconstruct those vertex

22 vectors from linear combinations of i significant principal components without losing much fidelity. part of PCA transformation matrix, or coefficients of linear combinations for reconstruction are used for input to the neuron network. From current experiments, with four out of thirty prinicipal components, the reconstruction error could be less than PCA on Texture Motion Ficial textures are also highly correlated, but PC analysis on texture is more complicated. In order to perform PC analysis, we first need to provide textures in same size. Therefore, we find the texture patch, the bounding box in the first neutral image from the projection of picked vertices. The patch, shown as pink pixel grid on the left side of Fig.8, is the image projection of partial mesh covered by picked vertices (red and blue points. We then select three vertices (red points on the left, whose projections are closest to the middle-top, middle-left and middle-right of the texture patch, and use them as a basis for warping or affine transformation. Fig. 8. Warping Due to non-rigid motions, the corresponding picked vertices in the following (m-1 frames might move, so do their image projections (see Fig.8, right. By aligning three

23 red-marked points on the right image with those on the left image (arrows, we can derive the matrices for warping or affine transformation against the first neutral frame. Next, for each pixel in the texture patch on the first frame image, we backward project it on the following (m-1 frames images, and assign interpolated values (RGB values to it. This way, we obtain (m-1 texture patches for the following frames. They are in the same size, but the figures could be warped. Finally, like PCA on vertex, we align each texture patch into a column vector, in the form of [,,,,,,...]. Then we apply PC analysis on these m vectors. Because these m vectors are also highly correlated, we also get a few significant principal components with MSE error less than part of PCA transformation matrix, or coefficients of linear combinations for reconstruction are used for target of the neuron network. 5.3 Neuron Network A neuron network can have multiple layers. Fig.9 shows a m layer network. Each layer has a weight matrix, a bias vertor. Transfer function at each layer may be different, so does the number of neurons,, which decides the size of the output,. Input to the first layer is the training input P. Output at the last layer is the training target T. Input vectors and the corresponding target vectors are used to train a network unitl it can approximate a function, associate input vectors with specific output vectors. Network with biases, a sigmoid layer and a linear output layer are capable of approximating any function with a finite number of discontinuities.

24 Fig. 9. Neuron Network 5.4 Motion Driven Textures We create both one-layer and two-layer neuron networks. Although the training performance is different, there are no much difference on simulation results. With the trained network, we provide motion vertices (reused, not new, the feedback are new textures, so-called motion driven textures. Fig.10 display a sequence of motion driven textures on the left eye corner area, which seems well matching the real textures. Note because of memory limitation, programs currently cannot generate bigger texture patches unless they are stitched together. Fig. 10. Animation on motion driven textures

25 6 CONCLUSIONS A list of work to do in the future is: do more experiments on different models, different expressions; view-dependent texture mapping; improve tracking; explore different algorithms to find non-rigid motions; adapt to PCA on large data set; build facial texture database; generate real-time animations. ACKNOWLEDGMENTS We would like to thank Trevor Chen, Michael M. Cohen and Rashid Clark at Experimental Psychology, UCSC for their help with data collection and suggestions. References [1] Volker Blanz and Thomas Vetter. A morphable model for the synthesis of 3d faces. Computer Graphics (ACM SIGGRAPH Proceedings, pages , August [2] Michael Garland. Qslim simplification software. garland/software/qslim.html. [3] Rafael Gonzalez and Richard Woods. Digital Image Processing. Addison-Wesley, [4] Zicheng liu, Ying Shan, and Zhengyou Zhang. Expressive expression mapping with ratio images. Computer Graphics (ACM SIGGRAPH Proceedings, [5] MATHWORK. Neural network toolbox, version 4. [6] F. Pighin, J. Hecker, D. Lischinski, R. Szeliski, and D.H. Salesin. Synthesizing realistic facial expressions from photographs. Computer Graphics (ACM SIGGRAPH Proceedings, pages 75 84, August 1998.

26 [7] Jun yong Noh and Ulrich Neumann. Jun-yong noh and ulrich neumann. Computer Graphics (ACM SIGGRAPH Proceedings, [8] Zhengyou Zhang. A flexible new technique for camera calibration. zhang/calib/.

Synthesizing Realistic Facial Expressions from Photographs

Synthesizing Realistic Facial Expressions from Photographs Synthesizing Realistic Facial Expressions from Photographs 1998 F. Pighin, J Hecker, D. Lischinskiy, R. Szeliskiz and D. H. Salesin University of Washington, The Hebrew University Microsoft Research 1

More information

A Morphable Model for the Synthesis of 3D Faces

A Morphable Model for the Synthesis of 3D Faces A Morphable Model for the Synthesis of 3D Faces Marco Nef Volker Blanz, Thomas Vetter SIGGRAPH 99, Los Angeles Presentation overview Motivation Introduction Database Morphable 3D Face Model Matching a

More information

Sample Based Texture extraction for Model based coding

Sample Based Texture extraction for Model based coding DEPARTMENT OF APPLIED PHYSICS AND ELECTRONICS UMEÅ UNIVERISTY, SWEDEN DIGITAL MEDIA LAB Sample Based Texture extraction for Model based coding Zhengrong Yao 1 Dept. Applied Physics and Electronics Umeå

More information

2D rendering takes a photo of the 2D scene with a virtual camera that selects an axis aligned rectangle from the scene. The photograph is placed into

2D rendering takes a photo of the 2D scene with a virtual camera that selects an axis aligned rectangle from the scene. The photograph is placed into 2D rendering takes a photo of the 2D scene with a virtual camera that selects an axis aligned rectangle from the scene. The photograph is placed into the viewport of the current application window. A pixel

More information

3D Geometry and Camera Calibration

3D Geometry and Camera Calibration 3D Geometry and Camera Calibration 3D Coordinate Systems Right-handed vs. left-handed x x y z z y 2D Coordinate Systems 3D Geometry Basics y axis up vs. y axis down Origin at center vs. corner Will often

More information

Speech Driven Synthesis of Talking Head Sequences

Speech Driven Synthesis of Talking Head Sequences 3D Image Analysis and Synthesis, pp. 5-56, Erlangen, November 997. Speech Driven Synthesis of Talking Head Sequences Peter Eisert, Subhasis Chaudhuri,andBerndGirod Telecommunications Laboratory, University

More information

Image-Based Deformation of Objects in Real Scenes

Image-Based Deformation of Objects in Real Scenes Image-Based Deformation of Objects in Real Scenes Han-Vit Chung and In-Kwon Lee Dept. of Computer Science, Yonsei University sharpguy@cs.yonsei.ac.kr, iklee@yonsei.ac.kr Abstract. We present a new method

More information

Abstract We present a system which automatically generates a 3D face model from a single frontal image of a face. Our system consists of two component

Abstract We present a system which automatically generates a 3D face model from a single frontal image of a face. Our system consists of two component A Fully Automatic System To Model Faces From a Single Image Zicheng Liu Microsoft Research August 2003 Technical Report MSR-TR-2003-55 Microsoft Research Microsoft Corporation One Microsoft Way Redmond,

More information

Computer Animation Visualization. Lecture 5. Facial animation

Computer Animation Visualization. Lecture 5. Facial animation Computer Animation Visualization Lecture 5 Facial animation Taku Komura Facial Animation The face is deformable Need to decide how all the vertices on the surface shall move Manually create them Muscle-based

More information

Faces. Face Modeling. Topics in Image-Based Modeling and Rendering CSE291 J00 Lecture 17

Faces. Face Modeling. Topics in Image-Based Modeling and Rendering CSE291 J00 Lecture 17 Face Modeling Topics in Image-Based Modeling and Rendering CSE291 J00 Lecture 17 Faces CS291-J00, Winter 2003 From David Romdhani Kriegman, slides 2003 1 Approaches 2-D Models morphing, indexing, etc.

More information

Image Morphing. Application: Movie Special Effects. Application: Registration /Alignment. Image Cross-Dissolve

Image Morphing. Application: Movie Special Effects. Application: Registration /Alignment. Image Cross-Dissolve Image Morphing Application: Movie Special Effects Morphing is turning one image into another (through a seamless transition) First movies with morphing Willow, 1988 Indiana Jones and the Last Crusade,

More information

EE795: Computer Vision and Intelligent Systems

EE795: Computer Vision and Intelligent Systems EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 14 130307 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Stereo Dense Motion Estimation Translational

More information

SYNTHESIS OF 3D FACES

SYNTHESIS OF 3D FACES SYNTHESIS OF 3D FACES R. Enciso, J. Li, D.A. Fidaleo, T-Y Kim, J-Y Noh and U. Neumann Integrated Media Systems Center University of Southern California Los Angeles, CA 90089, U.S.A. Abstract In this paper,

More information

calibrated coordinates Linear transformation pixel coordinates

calibrated coordinates Linear transformation pixel coordinates 1 calibrated coordinates Linear transformation pixel coordinates 2 Calibration with a rig Uncalibrated epipolar geometry Ambiguities in image formation Stratified reconstruction Autocalibration with partial

More information

Motion Synthesis and Editing. Yisheng Chen

Motion Synthesis and Editing. Yisheng Chen Motion Synthesis and Editing Yisheng Chen Overview Data driven motion synthesis automatically generate motion from a motion capture database, offline or interactive User inputs Large, high-dimensional

More information

FACIAL ANIMATION FROM SEVERAL IMAGES

FACIAL ANIMATION FROM SEVERAL IMAGES International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 5. Hakodate 1998 FACIAL ANIMATION FROM SEVERAL IMAGES Yasuhiro MUKAIGAWAt Yuichi NAKAMURA+ Yuichi OHTA+ t Department of Information

More information

Image-Based Rendering

Image-Based Rendering Image-Based Rendering COS 526, Fall 2016 Thomas Funkhouser Acknowledgments: Dan Aliaga, Marc Levoy, Szymon Rusinkiewicz What is Image-Based Rendering? Definition 1: the use of photographic imagery to overcome

More information

03 - Reconstruction. Acknowledgements: Olga Sorkine-Hornung. CSCI-GA Geometric Modeling - Spring 17 - Daniele Panozzo

03 - Reconstruction. Acknowledgements: Olga Sorkine-Hornung. CSCI-GA Geometric Modeling - Spring 17 - Daniele Panozzo 3 - Reconstruction Acknowledgements: Olga Sorkine-Hornung Geometry Acquisition Pipeline Scanning: results in range images Registration: bring all range images to one coordinate system Stitching/ reconstruction:

More information

3D FACE RECONSTRUCTION BASED ON EPIPOLAR GEOMETRY

3D FACE RECONSTRUCTION BASED ON EPIPOLAR GEOMETRY IJDW Volume 4 Number January-June 202 pp. 45-50 3D FACE RECONSRUCION BASED ON EPIPOLAR GEOMERY aher Khadhraoui, Faouzi Benzarti 2 and Hamid Amiri 3,2,3 Signal, Image Processing and Patterns Recognition

More information

CS 231. Deformation simulation (and faces)

CS 231. Deformation simulation (and faces) CS 231 Deformation simulation (and faces) Deformation BODY Simulation Discretization Spring-mass models difficult to model continuum properties Simple & fast to implement and understand Finite Element

More information

1. Mesh Coloring a.) Assign unique color to each polygon based on the polygon id.

1. Mesh Coloring a.) Assign unique color to each polygon based on the polygon id. 1. Mesh Coloring a.) Assign unique color to each polygon based on the polygon id. Figure 1: The dragon model is shown rendered using a coloring scheme based on coloring each triangle face according to

More information

Feature Tracking and Optical Flow

Feature Tracking and Optical Flow Feature Tracking and Optical Flow Prof. D. Stricker Doz. G. Bleser Many slides adapted from James Hays, Derek Hoeim, Lana Lazebnik, Silvio Saverse, who 1 in turn adapted slides from Steve Seitz, Rick Szeliski,

More information

CHAPTER 3 DISPARITY AND DEPTH MAP COMPUTATION

CHAPTER 3 DISPARITY AND DEPTH MAP COMPUTATION CHAPTER 3 DISPARITY AND DEPTH MAP COMPUTATION In this chapter we will discuss the process of disparity computation. It plays an important role in our caricature system because all 3D coordinates of nodes

More information

Hartley - Zisserman reading club. Part I: Hartley and Zisserman Appendix 6: Part II: Zhengyou Zhang: Presented by Daniel Fontijne

Hartley - Zisserman reading club. Part I: Hartley and Zisserman Appendix 6: Part II: Zhengyou Zhang: Presented by Daniel Fontijne Hartley - Zisserman reading club Part I: Hartley and Zisserman Appendix 6: Iterative estimation methods Part II: Zhengyou Zhang: A Flexible New Technique for Camera Calibration Presented by Daniel Fontijne

More information

Image Coding with Active Appearance Models

Image Coding with Active Appearance Models Image Coding with Active Appearance Models Simon Baker, Iain Matthews, and Jeff Schneider CMU-RI-TR-03-13 The Robotics Institute Carnegie Mellon University Abstract Image coding is the task of representing

More information

Structure from Motion

Structure from Motion 11/18/11 Structure from Motion Computer Vision CS 143, Brown James Hays Many slides adapted from Derek Hoiem, Lana Lazebnik, Silvio Saverese, Steve Seitz, and Martial Hebert This class: structure from

More information

CS 231. Deformation simulation (and faces)

CS 231. Deformation simulation (and faces) CS 231 Deformation simulation (and faces) 1 Cloth Simulation deformable surface model Represent cloth model as a triangular or rectangular grid Points of finite mass as vertices Forces or energies of points

More information

Animation. CS 4620 Lecture 33. Cornell CS4620 Fall Kavita Bala

Animation. CS 4620 Lecture 33. Cornell CS4620 Fall Kavita Bala Animation CS 4620 Lecture 33 Cornell CS4620 Fall 2015 1 Announcements Grading A5 (and A6) on Monday after TG 4621: one-on-one sessions with TA this Friday w/ prior instructor Steve Marschner 2 Quaternions

More information

3D Sensing and Reconstruction Readings: Ch 12: , Ch 13: ,

3D Sensing and Reconstruction Readings: Ch 12: , Ch 13: , 3D Sensing and Reconstruction Readings: Ch 12: 12.5-6, Ch 13: 13.1-3, 13.9.4 Perspective Geometry Camera Model Stereo Triangulation 3D Reconstruction by Space Carving 3D Shape from X means getting 3D coordinates

More information

PERFORMANCE CAPTURE FROM SPARSE MULTI-VIEW VIDEO

PERFORMANCE CAPTURE FROM SPARSE MULTI-VIEW VIDEO Stefan Krauß, Juliane Hüttl SE, SoSe 2011, HU-Berlin PERFORMANCE CAPTURE FROM SPARSE MULTI-VIEW VIDEO 1 Uses of Motion/Performance Capture movies games, virtual environments biomechanics, sports science,

More information

Animated Talking Head With Personalized 3D Head Model

Animated Talking Head With Personalized 3D Head Model Animated Talking Head With Personalized 3D Head Model L.S.Chen, T.S.Huang - Beckman Institute & CSL University of Illinois, Urbana, IL 61801, USA; lchen@ifp.uiuc.edu Jörn Ostermann, AT&T Labs-Research,

More information

NON-RIGID 3D SHAPE RECOVERY USING STEREO FACTORIZATION

NON-RIGID 3D SHAPE RECOVERY USING STEREO FACTORIZATION NON-IGID D SHAPE ECOVEY USING STEEO FACTOIZATION Alessio Del Bue ourdes Agapito Department of Computer Science Queen Mary, University of ondon ondon, E1 NS, UK falessio,lourdesg@dcsqmulacuk ABSTACT In

More information

Performance Driven Facial Animation using Blendshape Interpolation

Performance Driven Facial Animation using Blendshape Interpolation Performance Driven Facial Animation using Blendshape Interpolation Erika Chuang Chris Bregler Computer Science Department Stanford University Abstract This paper describes a method of creating facial animation

More information

BSB663 Image Processing Pinar Duygulu. Slides are adapted from Selim Aksoy

BSB663 Image Processing Pinar Duygulu. Slides are adapted from Selim Aksoy BSB663 Image Processing Pinar Duygulu Slides are adapted from Selim Aksoy Image matching Image matching is a fundamental aspect of many problems in computer vision. Object or scene recognition Solving

More information

On the Dimensionality of Deformable Face Models

On the Dimensionality of Deformable Face Models On the Dimensionality of Deformable Face Models CMU-RI-TR-06-12 Iain Matthews, Jing Xiao, and Simon Baker The Robotics Institute Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213 Abstract

More information

Il colore: acquisizione e visualizzazione. Lezione 17: 11 Maggio 2012

Il colore: acquisizione e visualizzazione. Lezione 17: 11 Maggio 2012 Il colore: acquisizione e visualizzazione Lezione 17: 11 Maggio 2012 The importance of color information Precision vs. Perception 3D scanned geometry Photo Color and appearance Pure geometry Pure color

More information

Facial Image Synthesis 1 Barry-John Theobald and Jeffrey F. Cohn

Facial Image Synthesis 1 Barry-John Theobald and Jeffrey F. Cohn Facial Image Synthesis Page 1 of 5 Facial Image Synthesis 1 Barry-John Theobald and Jeffrey F. Cohn 1 Introduction Facial expression has been central to the

More information

Radiosity. Johns Hopkins Department of Computer Science Course : Rendering Techniques, Professor: Jonathan Cohen

Radiosity. Johns Hopkins Department of Computer Science Course : Rendering Techniques, Professor: Jonathan Cohen Radiosity Radiosity Concept Global computation of diffuse interreflections among scene objects Diffuse lighting changes fairly slowly across a surface Break surfaces up into some number of patches Assume

More information

Reminder: Lecture 20: The Eight-Point Algorithm. Essential/Fundamental Matrix. E/F Matrix Summary. Computing F. Computing F from Point Matches

Reminder: Lecture 20: The Eight-Point Algorithm. Essential/Fundamental Matrix. E/F Matrix Summary. Computing F. Computing F from Point Matches Reminder: Lecture 20: The Eight-Point Algorithm F = -0.00310695-0.0025646 2.96584-0.028094-0.00771621 56.3813 13.1905-29.2007-9999.79 Readings T&V 7.3 and 7.4 Essential/Fundamental Matrix E/F Matrix Summary

More information

Free-Form Deformation and Other Deformation Techniques

Free-Form Deformation and Other Deformation Techniques Free-Form Deformation and Other Deformation Techniques Deformation Deformation Basic Definition Deformation: A transformation/mapping of the positions of every particle in the original object to those

More information

The Novel Approach for 3D Face Recognition Using Simple Preprocessing Method

The Novel Approach for 3D Face Recognition Using Simple Preprocessing Method The Novel Approach for 3D Face Recognition Using Simple Preprocessing Method Parvin Aminnejad 1, Ahmad Ayatollahi 2, Siamak Aminnejad 3, Reihaneh Asghari Abstract In this work, we presented a novel approach

More information

Face Tracking. Synonyms. Definition. Main Body Text. Amit K. Roy-Chowdhury and Yilei Xu. Facial Motion Estimation

Face Tracking. Synonyms. Definition. Main Body Text. Amit K. Roy-Chowdhury and Yilei Xu. Facial Motion Estimation Face Tracking Amit K. Roy-Chowdhury and Yilei Xu Department of Electrical Engineering, University of California, Riverside, CA 92521, USA {amitrc,yxu}@ee.ucr.edu Synonyms Facial Motion Estimation Definition

More information

CSE452 Computer Graphics

CSE452 Computer Graphics CSE452 Computer Graphics Lecture 19: From Morphing To Animation Capturing and Animating Skin Deformation in Human Motion, Park and Hodgins, SIGGRAPH 2006 CSE452 Lecture 19: From Morphing to Animation 1

More information

Stereo CSE 576. Ali Farhadi. Several slides from Larry Zitnick and Steve Seitz

Stereo CSE 576. Ali Farhadi. Several slides from Larry Zitnick and Steve Seitz Stereo CSE 576 Ali Farhadi Several slides from Larry Zitnick and Steve Seitz Why do we perceive depth? What do humans use as depth cues? Motion Convergence When watching an object close to us, our eyes

More information

Muscle Based facial Modeling. Wei Xu

Muscle Based facial Modeling. Wei Xu Muscle Based facial Modeling Wei Xu Facial Modeling Techniques Facial modeling/animation Geometry manipulations Interpolation Parameterizations finite element methods muscle based modeling visual simulation

More information

3D Modeling of Objects Using Laser Scanning

3D Modeling of Objects Using Laser Scanning 1 3D Modeling of Objects Using Laser Scanning D. Jaya Deepu, LPU University, Punjab, India Email: Jaideepudadi@gmail.com Abstract: In the last few decades, constructing accurate three-dimensional models

More information

Faces and Image-Based Lighting

Faces and Image-Based Lighting Announcements Faces and Image-Based Lighting Project #3 artifacts voting Final project: Demo on 6/25 (Wednesday) 13:30pm in this room Reports and videos due on 6/26 (Thursday) 11:59pm Digital Visual Effects,

More information

Image Registration Lecture 4: First Examples

Image Registration Lecture 4: First Examples Image Registration Lecture 4: First Examples Prof. Charlene Tsai Outline Example Intensity-based registration SSD error function Image mapping Function minimization: Gradient descent Derivative calculation

More information

Index. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 253

Index. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 253 Index 3D reconstruction, 123 5+1-point algorithm, 274 5-point algorithm, 260 7-point algorithm, 255 8-point algorithm, 253 affine point, 43 affine transformation, 55 affine transformation group, 55 affine

More information

Final Review CMSC 733 Fall 2014

Final Review CMSC 733 Fall 2014 Final Review CMSC 733 Fall 2014 We have covered a lot of material in this course. One way to organize this material is around a set of key equations and algorithms. You should be familiar with all of these,

More information

SCAPE: Shape Completion and Animation of People

SCAPE: Shape Completion and Animation of People SCAPE: Shape Completion and Animation of People By Dragomir Anguelov, Praveen Srinivasan, Daphne Koller, Sebastian Thrun, Jim Rodgers, James Davis From SIGGRAPH 2005 Presentation for CS468 by Emilio Antúnez

More information

Motion Estimation. There are three main types (or applications) of motion estimation:

Motion Estimation. There are three main types (or applications) of motion estimation: Members: D91922016 朱威達 R93922010 林聖凱 R93922044 謝俊瑋 Motion Estimation There are three main types (or applications) of motion estimation: Parametric motion (image alignment) The main idea of parametric motion

More information

Topics and things to know about them:

Topics and things to know about them: Practice Final CMSC 427 Distributed Tuesday, December 11, 2007 Review Session, Monday, December 17, 5:00pm, 4424 AV Williams Final: 10:30 AM Wednesday, December 19, 2007 General Guidelines: The final will

More information

Processing 3D Surface Data

Processing 3D Surface Data Processing 3D Surface Data Computer Animation and Visualisation Lecture 12 Institute for Perception, Action & Behaviour School of Informatics 3D Surfaces 1 3D surface data... where from? Iso-surfacing

More information

Geometric Transformations and Image Warping

Geometric Transformations and Image Warping Geometric Transformations and Image Warping Ross Whitaker SCI Institute, School of Computing University of Utah Univ of Utah, CS6640 2009 1 Geometric Transformations Greyscale transformations -> operate

More information

The exam begins at 2:40pm and ends at 4:00pm. You must turn your exam in when time is announced or risk not having it accepted.

The exam begins at 2:40pm and ends at 4:00pm. You must turn your exam in when time is announced or risk not having it accepted. CS 184: Foundations of Computer Graphics page 1 of 10 Student Name: Class Account Username: Instructions: Read them carefully! The exam begins at 2:40pm and ends at 4:00pm. You must turn your exam in when

More information

Index. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 263

Index. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 263 Index 3D reconstruction, 125 5+1-point algorithm, 284 5-point algorithm, 270 7-point algorithm, 265 8-point algorithm, 263 affine point, 45 affine transformation, 57 affine transformation group, 57 affine

More information

COMPUTER AND ROBOT VISION

COMPUTER AND ROBOT VISION VOLUME COMPUTER AND ROBOT VISION Robert M. Haralick University of Washington Linda G. Shapiro University of Washington T V ADDISON-WESLEY PUBLISHING COMPANY Reading, Massachusetts Menlo Park, California

More information

Rigid ICP registration with Kinect

Rigid ICP registration with Kinect Rigid ICP registration with Kinect Students: Yoni Choukroun, Elie Semmel Advisor: Yonathan Aflalo 1 Overview.p.3 Development of the project..p.3 Papers p.4 Project algorithm..p.6 Result of the whole body.p.7

More information

Accurate Reconstruction by Interpolation

Accurate Reconstruction by Interpolation Accurate Reconstruction by Interpolation Leow Wee Kheng Department of Computer Science School of Computing National University of Singapore International Conference on Inverse Problems and Related Topics

More information

Stereo and Epipolar geometry

Stereo and Epipolar geometry Previously Image Primitives (feature points, lines, contours) Today: Stereo and Epipolar geometry How to match primitives between two (multiple) views) Goals: 3D reconstruction, recognition Jana Kosecka

More information

Transforming Objects and Components

Transforming Objects and Components 4 Transforming Objects and Components Arrow selection Lasso selection Paint selection Move Rotate Scale Universal Manipulator Soft Modification Show Manipulator Last tool used Figure 4.1 Maya s manipulation

More information

3D Sensing. 3D Shape from X. Perspective Geometry. Camera Model. Camera Calibration. General Stereo Triangulation.

3D Sensing. 3D Shape from X. Perspective Geometry. Camera Model. Camera Calibration. General Stereo Triangulation. 3D Sensing 3D Shape from X Perspective Geometry Camera Model Camera Calibration General Stereo Triangulation 3D Reconstruction 3D Shape from X shading silhouette texture stereo light striping motion mainly

More information

Geometric camera models and calibration

Geometric camera models and calibration Geometric camera models and calibration http://graphics.cs.cmu.edu/courses/15-463 15-463, 15-663, 15-862 Computational Photography Fall 2018, Lecture 13 Course announcements Homework 3 is out. - Due October

More information

Statistical Learning of Human Body through Feature Wireframe

Statistical Learning of Human Body through Feature Wireframe Statistical Learning of Human Body through Feature Wireframe Jida HUANG 1, Tsz-Ho KWOK 2*, Chi ZHOU 1 1 Industrial and Systems Engineering, University at Buffalo, SUNY, Buffalo NY, USA; 2 Mechanical, Industrial

More information

Il colore: acquisizione e visualizzazione. Lezione 20: 11 Maggio 2011

Il colore: acquisizione e visualizzazione. Lezione 20: 11 Maggio 2011 Il colore: acquisizione e visualizzazione Lezione 20: 11 Maggio 2011 Outline The importance of color What is color? Material properties vs. unshaded color Texture building from photos Image registration

More information

Stereo Vision. MAN-522 Computer Vision

Stereo Vision. MAN-522 Computer Vision Stereo Vision MAN-522 Computer Vision What is the goal of stereo vision? The recovery of the 3D structure of a scene using two or more images of the 3D scene, each acquired from a different viewpoint in

More information

Why animate humans? Why is this hard? Aspects of the Problem. These lectures. Animation Apreciation 101

Why animate humans? Why is this hard? Aspects of the Problem. These lectures. Animation Apreciation 101 Animation by Example Lecture 1: Introduction, Human Representation Michael Gleicher University of Wisconsin- Madison www.cs.wisc.edu/~gleicher www.cs.wisc.edu/graphics Why animate humans? Movies Television

More information

S U N G - E U I YO O N, K A I S T R E N D E R I N G F R E E LY A VA I L A B L E O N T H E I N T E R N E T

S U N G - E U I YO O N, K A I S T R E N D E R I N G F R E E LY A VA I L A B L E O N T H E I N T E R N E T S U N G - E U I YO O N, K A I S T R E N D E R I N G F R E E LY A VA I L A B L E O N T H E I N T E R N E T Copyright 2018 Sung-eui Yoon, KAIST freely available on the internet http://sglab.kaist.ac.kr/~sungeui/render

More information

Camera Calibration. COS 429 Princeton University

Camera Calibration. COS 429 Princeton University Camera Calibration COS 429 Princeton University Point Correspondences What can you figure out from point correspondences? Noah Snavely Point Correspondences X 1 X 4 X 3 X 2 X 5 X 6 X 7 p 1,1 p 1,2 p 1,3

More information

BCC Rays Ripply Filter

BCC Rays Ripply Filter BCC Rays Ripply Filter The BCC Rays Ripply filter combines a light rays effect with a rippled light effect. The resulting light is generated from a selected channel in the source image and spreads from

More information

ENGN2911I: 3D Photography and Geometry Processing Assignment 1: 3D Photography using Planar Shadows

ENGN2911I: 3D Photography and Geometry Processing Assignment 1: 3D Photography using Planar Shadows ENGN2911I: 3D Photography and Geometry Processing Assignment 1: 3D Photography using Planar Shadows Instructor: Gabriel Taubin Assignment written by: Douglas Lanman 29 January 2009 Figure 1: 3D Photography

More information

Registration of Expressions Data using a 3D Morphable Model

Registration of Expressions Data using a 3D Morphable Model Registration of Expressions Data using a 3D Morphable Model Curzio Basso, Pascal Paysan, Thomas Vetter Computer Science Department, University of Basel {curzio.basso,pascal.paysan,thomas.vetter}@unibas.ch

More information

Matching. Compare region of image to region of image. Today, simplest kind of matching. Intensities similar.

Matching. Compare region of image to region of image. Today, simplest kind of matching. Intensities similar. Matching Compare region of image to region of image. We talked about this for stereo. Important for motion. Epipolar constraint unknown. But motion small. Recognition Find object in image. Recognize object.

More information

Human body animation. Computer Animation. Human Body Animation. Skeletal Animation

Human body animation. Computer Animation. Human Body Animation. Skeletal Animation Computer Animation Aitor Rovira March 2010 Human body animation Based on slides by Marco Gillies Human Body Animation Skeletal Animation Skeletal Animation (FK, IK) Motion Capture Motion Editing (retargeting,

More information

Accurate 3D Face and Body Modeling from a Single Fixed Kinect

Accurate 3D Face and Body Modeling from a Single Fixed Kinect Accurate 3D Face and Body Modeling from a Single Fixed Kinect Ruizhe Wang*, Matthias Hernandez*, Jongmoo Choi, Gérard Medioni Computer Vision Lab, IRIS University of Southern California Abstract In this

More information

FACIAL ANIMATION WITH MOTION CAPTURE BASED ON SURFACE BLENDING

FACIAL ANIMATION WITH MOTION CAPTURE BASED ON SURFACE BLENDING FACIAL ANIMATION WITH MOTION CAPTURE BASED ON SURFACE BLENDING Lijia Zhu and Won-Sook Lee School of Information Technology and Engineering, University of Ottawa 800 King Edward Ave., Ottawa, Ontario, Canada,

More information

2D/3D Geometric Transformations and Scene Graphs

2D/3D Geometric Transformations and Scene Graphs 2D/3D Geometric Transformations and Scene Graphs Week 4 Acknowledgement: The course slides are adapted from the slides prepared by Steve Marschner of Cornell University 1 A little quick math background

More information

MERGING POINT CLOUDS FROM MULTIPLE KINECTS. Nishant Rai 13th July, 2016 CARIS Lab University of British Columbia

MERGING POINT CLOUDS FROM MULTIPLE KINECTS. Nishant Rai 13th July, 2016 CARIS Lab University of British Columbia MERGING POINT CLOUDS FROM MULTIPLE KINECTS Nishant Rai 13th July, 2016 CARIS Lab University of British Columbia Introduction What do we want to do? : Use information (point clouds) from multiple (2+) Kinects

More information

Learning a generic 3D face model from 2D image databases using incremental structure from motion

Learning a generic 3D face model from 2D image databases using incremental structure from motion Learning a generic 3D face model from 2D image databases using incremental structure from motion Jose Gonzalez-Mora 1,, Fernando De la Torre b, Nicolas Guil 1,, Emilio L. Zapata 1 a Department of Computer

More information

Linear Methods for Regression and Shrinkage Methods

Linear Methods for Regression and Shrinkage Methods Linear Methods for Regression and Shrinkage Methods Reference: The Elements of Statistical Learning, by T. Hastie, R. Tibshirani, J. Friedman, Springer 1 Linear Regression Models Least Squares Input vectors

More information

LUMS Mine Detector Project

LUMS Mine Detector Project LUMS Mine Detector Project Using visual information to control a robot (Hutchinson et al. 1996). Vision may or may not be used in the feedback loop. Visual (image based) features such as points, lines

More information

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the third chapter of the online book by Michael Nielson: neuralnetworksanddeeplearning.com

More information

CS6670: Computer Vision

CS6670: Computer Vision CS6670: Computer Vision Noah Snavely Lecture 7: Image Alignment and Panoramas What s inside your fridge? http://www.cs.washington.edu/education/courses/cse590ss/01wi/ Projection matrix intrinsics projection

More information

1 (5 max) 2 (10 max) 3 (20 max) 4 (30 max) 5 (10 max) 6 (15 extra max) total (75 max + 15 extra)

1 (5 max) 2 (10 max) 3 (20 max) 4 (30 max) 5 (10 max) 6 (15 extra max) total (75 max + 15 extra) Mierm Exam CS223b Stanford CS223b Computer Vision, Winter 2004 Feb. 18, 2004 Full Name: Email: This exam has 7 pages. Make sure your exam is not missing any sheets, and write your name on every page. The

More information

coding of various parts showing different features, the possibility of rotation or of hiding covering parts of the object's surface to gain an insight

coding of various parts showing different features, the possibility of rotation or of hiding covering parts of the object's surface to gain an insight Three-Dimensional Object Reconstruction from Layered Spatial Data Michael Dangl and Robert Sablatnig Vienna University of Technology, Institute of Computer Aided Automation, Pattern Recognition and Image

More information

Chapter 5. Transforming Shapes

Chapter 5. Transforming Shapes Chapter 5 Transforming Shapes It is difficult to walk through daily life without being able to see geometric transformations in your surroundings. Notice how the leaves of plants, for example, are almost

More information

Processing 3D Surface Data

Processing 3D Surface Data Processing 3D Surface Data Computer Animation and Visualisation Lecture 17 Institute for Perception, Action & Behaviour School of Informatics 3D Surfaces 1 3D surface data... where from? Iso-surfacing

More information

Animation. CS 465 Lecture 22

Animation. CS 465 Lecture 22 Animation CS 465 Lecture 22 Animation Industry production process leading up to animation What animation is How animation works (very generally) Artistic process of animation Further topics in how it works

More information

Lecture 7: Image Morphing. Idea #2: Align, then cross-disolve. Dog Averaging. Averaging vectors. Idea #1: Cross-Dissolving / Cross-fading

Lecture 7: Image Morphing. Idea #2: Align, then cross-disolve. Dog Averaging. Averaging vectors. Idea #1: Cross-Dissolving / Cross-fading Lecture 7: Image Morphing Averaging vectors v = p + α (q p) = (1 - α) p + α q where α = q - v p α v (1-α) q p and q can be anything: points on a plane (2D) or in space (3D) Colors in RGB or HSV (3D) Whole

More information

Acquisition and Visualization of Colored 3D Objects

Acquisition and Visualization of Colored 3D Objects Acquisition and Visualization of Colored 3D Objects Kari Pulli Stanford University Stanford, CA, U.S.A kapu@cs.stanford.edu Habib Abi-Rached, Tom Duchamp, Linda G. Shapiro and Werner Stuetzle University

More information

Interactive Deformation with Triangles

Interactive Deformation with Triangles Interactive Deformation with Triangles James Dean Palmer and Ergun Akleman Visualization Sciences Program Texas A&M University Jianer Chen Department of Computer Science Texas A&M University Abstract In

More information

TEXTURE OVERLAY ONTO NON-RIGID SURFACE USING COMMODITY DEPTH CAMERA

TEXTURE OVERLAY ONTO NON-RIGID SURFACE USING COMMODITY DEPTH CAMERA TEXTURE OVERLAY ONTO NON-RIGID SURFACE USING COMMODITY DEPTH CAMERA Tomoki Hayashi 1, Francois de Sorbier 1 and Hideo Saito 1 1 Graduate School of Science and Technology, Keio University, 3-14-1 Hiyoshi,

More information

Structure from Motion and Multi- view Geometry. Last lecture

Structure from Motion and Multi- view Geometry. Last lecture Structure from Motion and Multi- view Geometry Topics in Image-Based Modeling and Rendering CSE291 J00 Lecture 5 Last lecture S. J. Gortler, R. Grzeszczuk, R. Szeliski,M. F. Cohen The Lumigraph, SIGGRAPH,

More information

Progressive Mesh. Reddy Sambavaram Insomniac Games

Progressive Mesh. Reddy Sambavaram Insomniac Games Progressive Mesh Reddy Sambavaram Insomniac Games LOD Schemes Artist made LODs (time consuming, old but effective way) ViewDependentMesh (usually used for very large complicated meshes. CAD apps. Probably

More information

HIGH-RESOLUTION ANIMATION OF FACIAL DYNAMICS

HIGH-RESOLUTION ANIMATION OF FACIAL DYNAMICS HIGH-RESOLUTION ANIMATION OF FACIAL DYNAMICS N. Nadtoka, J.R. Tena, A. Hilton, J. Edge Centre for Vision, Speech and Signal Processing, University of Surrey {N.Nadtoka, J.Tena, A.Hilton}@surrey.ac.uk Keywords:

More information

Motion Interpretation and Synthesis by ICA

Motion Interpretation and Synthesis by ICA Motion Interpretation and Synthesis by ICA Renqiang Min Department of Computer Science, University of Toronto, 1 King s College Road, Toronto, ON M5S3G4, Canada Abstract. It is known that high-dimensional

More information

All the Polygons You Can Eat. Doug Rogers Developer Relations

All the Polygons You Can Eat. Doug Rogers Developer Relations All the Polygons You Can Eat Doug Rogers Developer Relations doug@nvidia.com Future of Games Very high resolution models 20,000 triangles per model Lots of them Complex Lighting Equations Floating point

More information

Sculpting 3D Models. Glossary

Sculpting 3D Models. Glossary A Array An array clones copies of an object in a pattern, such as in rows and columns, or in a circle. Each object in an array can be transformed individually. Array Flyout Array flyout is available in

More information

Vision-based Control of 3D Facial Animation

Vision-based Control of 3D Facial Animation Eurographics/SIGGRAPH Symposium on Computer Animation (2003) D. Breen, M. Lin (Editors) Vision-based Control of 3D Facial Animation Jin-xiang Chai,1 Jing Xiao1 and Jessica Hodgins1 1 The Robotics Institute,

More information