Detection

Project delveloped for the purpose of Computer Vision [CS418] course assignment 3 for Technical Univetsity Of Crete.
In this assignment we develop an object detector based on gradient features and sliding window classification.

Image Gradient

A function that takes a grayscale image as input and returns two arrays the same size as the image, the first of which contains the magnitude of the image gradient at each pixel and the second containing the orientation.

[mag,ori] = mygradient(I)
The function filters the image with the simple x- and y-derivative filters.

• Given the derivatives we can compute the orientation and magnitude of the gradient vector at each pixel.
• Used imfilter with the 'replicate' option in order to nicely handle the image boundaries.
• Visualized the resulting gradient magnitude and orientation using imagesc.
• Used colorbar and a non-grayscale colormap in order to make the visualization better.
• Used the colormap jet for the gradient magnitude, and colormap hsv for the orientation values.

Histograms of Gradient Orientations

A function that computes gradient orientation histograms over each 8x8 block of pixels. The function bins the orientation into 9 equal sized bins between -pi/2 and pi/2. The input of the function is an image of size HxW. The output is a three-dimensional array ohist whose size is (H/8)x(W/8)x9 where ohist(i,j,k) contains the count of how many edges of orientation k fell in block (i,j). To determine if a pixel is an edge, we need to choose some threshold. using a threshold that is a tenth the maximum gradient magnitude in the image (i.e. thresh = 0.1*max(mag(:))). Since each 8x8 block will contain a different number of edges, the resulting histogram gets normalized for each block to sum to 1 (i.e., sum(ohist,3) should be 1 at every block location). The function loops over the orientation bins. For each orientation it identifies those pixels in the image whose magnitude is above the threshold and whose orientation falls in the given bin. This is done using logical operations in order to generate an array the same size as the image that contains 1s at the locations of every edge pixel that falls in the given orientation bin and is above threshold. To collect up pixels in each 8x8 spatial block, the function im2col(..,[8 8],'distinct') is used. The im2col function will automatically pad out the image to a multiple of 8 which is convenient.

Detection

A function that takes a template (see the step 4) and an image and returns the top detections found in the image. It has the prototype [x,y,score] = detect(I,template,ndet) where ndet is the number of detections to return.

• First it computes the histogram-of-gradient-orientation feature map for the image, then correlates the template with the feature map. Since the feature map and template are both three dimensional, each orientation gets filtered separately and then the results summed up give the final response. This final response map will be of size (H/8)x(W/8).
• Implementation of non-maxima suppression. Done by sorting the responses in descending order of their score.
• Every time a detection is added to the list to return, computes overlap by calculating the distance between a pair of detections and checking that the distance is greater than say 70% of the width of the template to make sure that the location of this detection is not too close to any of the detections already in the output list.
• Returns the locations of the detections in terms of the original image pixel coordinates.

Detection Script

The provided demo detection script loads in a training image and a test image and has the user click on one or more patches of the training image. The script then builds an average template using the histogram feature map. Finally the script calls your detect function using this average template to detect objects in the test image.

Additional Features

Modified the provided demo detection script to add two additional features.

• First, making the width and height of the template a variable parameter so that it can be easily changed to detect different size objects. Specifying the template size in blocks (e.g. the provided script assumes the template is 16x16 blocks). This required a handful of changes to values that are hard-coded in the demo script.
• Second, modified the script to incorporate negative training examples and produce a final template which is the difference of positive and negative averages. Negative examples can either be generated by having the user click or by providing a negative training image which contains no instances of the object and sampling random locations.

Results

1. mygradient.m output

2. Demo Detector Output

2.1. Train image 1

2.2. Patches

2.3. Detection

2.4. Train Image 2

2.5. Patch

2.6. Detection

3. Modified Detector Output

3.1. Train Image

3.2. Templates

3.3. Detection