Hi, I am working on abnormal activity detection. I have image frames which represent various activities of person. I want to extract the shape descriptors from these images.I don't understand how to do that. Can someone please help me? I have attached the image from which I have to extract the features. How can I achieve that? Please help.
Hi Yashaswini,
If I were you, I would consider to define some kind of silhouette recognition pattern based on a shape context. I don't know if you need to define any 'shape descriptor', but you would need to precise what form/thing you want to describe. I mean, will your data do contain complex shapes or always the same? I guess that the first step of your work would be to extract the interest shape (volume-of-interest, VOI), then to segment and normalize the VOI to a common space or to define some parts in your VOI, which would represent your specific regions of interest (i.e. face, upper-limbs) and in which you could extract some features, such as volume or any index to quantify that part. I think that it would be a first idea of the concept. I attach you an example of shape-based hand recognition.
Article Shape Matching and Object Recognition Using Shape Contexts
Article Shape-based hand recognition
There are many that could be googled; you could kindly our publications based on mammography and CAD based medical image segmentation
Hello you can extract shape features within MATLAB tool. I have been extracting features like Area,Filled Area,Perimeter,Circularity,Compactness,Convexity,Roundness.for that I have been used regionprops() function in MATLAB.
Hello, Yashaswini.
You can use standard algorithms after Canny: OTSU’S method, Hough transform for edge linking, Edge linking using local processing. (You can find it in Gonzalez)
For example, in OpenCV recearchers used Suzuki and Abe algorithm. ( https://stackoverflow.com/questions/10427474/what-is-the-algorithm-that-opencv-uses-for-finding-contours).
- Badges
- Science topic
- Similar topics
- Engineering
- Computer Engineering