Stereo Image Analysis

This paper describes the development of
algorithms for moment-based stereopsis as the feature
descriptors and stereo matching algorithms. Moment
functions capture global characteristics of an image shape
and are ideally suited for obtaining the optimal matching
positions of small windowed regions in a stereo image pair.
Among the class of moment functions, discrete orthogonal
moments do not exhibit large dynamic range variations, are
robust with respect to image noise, and have superior
feature representation capabilities. These considerations
have led to the choice of using Scaled Tchebichef Moments as
feature descriptors for stereo analysis in this research.
The journal also compares the stereo matching performance of
conventional methods such as the cooperative stereopsis,
correlation and window-based matching techniques, with
Geometric and Tchebichef Moments. Extensive analysis using
various types of images (synthetic and real, binary and
gray-level) was carried out with interesting results. A
suitably chosen moment vector (known as Scaled Tchebichef
Moments) together with dynamic programming yielded highly
satisfactory results in a stereo matching algorithm.