TEXTURED IMAGE SEGMENTATION VIA SCATTERING TRANSFORM AND SCORE FUSION

This paper deals with the problem of textured image segmentation. Our approach employs a recently developed method, the so called Wavelet Scattering Transform, for the process of feature extraction to get robust and invariant system to small deformations. The classification of scattering energy distribution is done by support vector machine classifier(SVM). To obtain a more accurate result of segmentation, we integrated information fusion on score level to combine the scores obtained by SVM classifier during classification process, using probability theory. Experimental results demonstrated the effectiveness of the proposed approach and show that high accuracy can be achieved.