Methods for Detecting of Structural Changes in Computer Vision Systems

The automation of experimental investigations based on video recording and different artificial vision applications often require that changes in a sequence of frames be detected without the observer’s assistance. Variations in brightness, color, and size of an object are easily detectable using energy criteria. Nevertheless, some problems demand the use of algorithms capable of responding to small scale and texture changes of images. These problems can be solved by applying the criteria of Mean Structural Similarity Index Measure (MSSIM) and the developed Mean Nonparametric Structural Similarity Index Measure (MNSSIM), as well as the spectral algorithm for detecting structural changes in a frame, which have been used to good effect in video codec analysis. The profitable features of these criteria are their computational simplicity and their conformance to the human visual system. The criteria have not only a sensitivity for difference of comparing frames, but also have high stability of Gaussian and non-Guassian (impulse) noises. This chapter describes the MSSIM, the own developed MNSSIM algorithms, and the spectral criterion, which provides the experimental confirmation of operating characteristics and features. The use of these criteria in automatic detection of changes in video captured scientific research scenes, the detection of motion or variable fragments in video frames in the intelligent video systems, and the application in video coding systems are discussed.