Merged MPEG-7 visual descriptors for image classification

Abstract

In this paper we investigated how merging of MPEG-7 visual descriptors can improve accuracy of image classification systems. Combination of MPEG-7 Colour Layout Descriptor (CLD) and Edge Histogram Descriptor (EHD) was used to extract features for input into multiclass Support Vector Machine (SVM) classifier. SVM classifier was trained and evaluated on a standard test database of 1000 images classified into 10 categories. Achieved accuracy for merged descriptors has shown significant improvement compared to the results when CLD and EHD descriptors were used independently.