期刊名称：International Journal of Engineering and Computer Science
印刷版ISSN：2319-7242
出版年度：2013
卷号：2
期号：12
页码：3472-3476
出版社：IJECS
摘要：Interest point detection is the vital aspect of computer vision and image processing. Interest point detection is very importantfor image retrieval and object categorization systems from which local descriptors are calculated for image matching schemes. Earlierapproaches largely ignored color aspect as interest point calculation are largely based on luminance. However the use of color increases thedistinctiveness of interest points. Subsequently an approach that uses saliency-based feature selection aided with a principle componentanalysis-based scale selection method was developed that happens to be a light-invariant interest points detection system. This paperintroduces color interest points for sparse image representation. In the domain of video-indexing framework, interest points provides moreuseful information when compared to static images. Since human perception system is naturally influenced by motion of objects, we proposeto extend the above approach for dynamic video streams using Space-Time Interest Points (STIP) method. The method includes the processof calculating the interest points in 3D domain (i.e., for feature extraction, this means that the main 2D concepts for images are extended to3D). STIP renders moving objects in a live feed and characterizes the specific changes in the movement of these objects. A practicalimplementation of the proposed system validates our claim to support live video feeds and further it can be used in domains such as MotionTracking, Entity Detection and Naming applications that have abundance importance
