Are there any computer vision algorithms specifically targeting depth images?

What are the basic computer vision algorithms to start with for a beginner, which can be implemented as JAVA project?

• I am just graduated and confused about my majors for MS. I just wanna get a feel of computer vision algorithm, so i want to start implementing some algorithm. I have a basic knowledge of SIFT and RANSAC, but have no clue about implementation details. Can anyone guide me, with which algorithms should i start and what can be a basic project to start with in computer vision?          I was thinking of a project like a simple object path tracking or converting series of images into panaroma view. Can anyone help me what can be a basic one month project for a beginner with insight of computer vision?

• Answer:

  1.   Implementing Harris Corner Detection would be a good choice to begin with,  as it deals you with a lot matrix theory, feature extraction mechanisms, convolution theorem and all other good stuff to get you familiar with basic Computer Vision concepts. Instead of using readily available tools, one should compute all gradients, eigenvalues himself to get into the notion of how a feature detector works. This magical algorithm has given a big inspiration for further research in the field. 2.   Then go for Shi & Tommasi feature detection which mostly relies on Harris but reduces the computation cost mostly and gives better results with some images and videos. 3.   After Shi & Tommasi, you can look for Shi & Tommasi feature tracker, which detects a Shi & Tommasi corner in the first frame and tracks them all in subsequent frames. This will be funnier than above 2 as you will find yourself in front of the webcam swinging when evaluating your implementation. Check Lukas & Kanade tracker. 4.  Simple object tracking requires to detect a simple object and track it in subsequent frames which now you are familiar with. You can develop a blob detector & tracker by imitating regionProps command in Matlab. You can detect  a circle like area -which is the green ball you use for example - in the frame and track it by template matching.