A. Zizzari, B. Michaelis, and G. Gademann (Germany)
: Pattern Recognition, Co-occurrence Matrix, Textural Features, Geometrical Modelling.
The automatic detection of particular objects, chara cterized by specific textures in digital images, is usually obtained employing a standard classifier. It firstly discri minates among different feature values associated to different textures in the same image, and then it assigns these values to some respective belonging classes. Thus, the process of extracting the feature values is a very critical task, affecting the final level of recognition in the same classifier. Here, some optimal feature functions are introduced. These functions are computed on their respective co occurrence matrices associated to the image blocks, and are characterized in: increasing the distance between the distributions of feature values and, at the same time, decreasing the width of possible regions of overlapping. That leads to a significant improvement in the “level of separation” of the classes associated to the source infor mation and, consequently, to a significant improvement in the global classifier performance.
Important Links:
Go Back
