Lab 11. Basic Image Processing Algorithms Fall 2017

Transcription

1 Lab 11 Basic Image Processing Algorithms Fall 2017

2 Lab 11: video segmentation with temporal histogram script: function: loads in a video file --- it will be a 4D array in the MATLAB Workspace (stacked RGB images) calls your function saves your results into another video file --- plotting the results every timestep to a figure, and saving it as a frame of your output video input: stacked RGB images (4D array) (+ 2 technical parameters) output: 5 arrays --- results of the different processing steps of the video segmentation Deadline to upload to the SVN server: 23:59, 12/12/2017 2

3 3

4 Step 1: create a function inputs: 4D array of color images (H x W x Ch x frame_number) length of running window for temporal background comp. (scalar, let's have the value 100) threshold value for grayscale bw conversion (scalar, try value 50 in the first run) outputs - all of them 3D arrays: array of actual grayscale images array of background images array of differences array of thresholded differences array of morphologically processed difference-images convert your 4D array to a 3D array of grayscale images with type double you can do it only frame-by-frame! due to the huge size of the input video, for your initial experiments define a desired_length of 30 frames allocate space for all of your output arrays (with the desired_length) fill the elements of your first returned array (array of actual grayscale images) appropriately 4

5 Step 2: create a script load the input video file in MATLAB, you will need a VideoReader object: video_in = VideoReader( video1.avi ); % under MAC/linux: video1.ogv suggested the read operation on this object returns you the image-sequence as a 4D array (please check the size of it, which should be equal to [240x320x3x1001]) call your function open the output video file in MATLAB, you will need a VideoWriter object: video_out = VideoWriter( video1_your_output.avi ); (if you can t see the output later, try to set the format to Uncompressed AVI but be careful, it will result you a huge file!) in the case of your video writer: you can specify the framerate only before the opening: video_out.framerate = video_in.framerate; the open operation on this object opens you the writer do not forget to call close operation on this object at the end of your script: close(video_out); 5

6 Step 2: create a script - continued initialize a new figure with 6 subplots (2x3) iterate over the temporal dimension of the first returned value of your function (array of actual grayscale images): in the different subplots plot the appropriate slice of your different output arrays (converted to uint8, of course) at the end of your loop body, call the writevideo method on your output video object, with the current frame as a second parameter: writevideo(video_out, getframe(gcf)); 6

7 Step 3: extend your function Fill all of your arrays in the same for-loop as follows: 1. array of actual grayscale images: simply a frame belonging to the actual iteration 2. array of background images: the statistical background in a predefined time-window (as the 2nd parameter of your function indicates; please compute it backward from the current iteration); with the built-in mode function you can compute the mode of an array along a specified dimension, eg. temporal_background = mode(grayscale_video(:, :, idx-t:idx-1), 3); 3. array of differences: the difference (positive number! --- use abs) between the actual grayscale image and the background image 4. array of thresholded differences: the b&w version of the difference (computed with 3rd param) think about the particularly excellent array indexing capabilities of MATLAB, like data_array = [1, 2, 3; 4, 5, 6]; threshold_value = 3; data_array_copy = 1.* data_array; data_array_copy(data_array_copy > threshold_value) = 6; data_array_copy(data_array_copy <= threshold_value) = 1; 5. array of morphologically processed difference-images: apply the morphological opening (erosion + dilation) on your thresholded diff. image, useful MATLAB commands: imerode, imdilate both of them will need a structuring element (strel) as a second parameter (it can be disk-shaped with radius 1, eg strel( disk, 1) --- theory recap on the next slides 7

8 8

9 9

10 10