Feature Extraction Framework for CBIR Systems based on Cyclic Transform Analysis & Spatial Information

A novel framework of feature generation for Content based image retrieval (CBIR) is proposed. This system is realized on Cyclic transform Analysis (CTA). It introduces statistical descriptors in the signals frequency domain. Then the CT of data is computed by Semi supervised algorithm (SSA) which is a simple & efficient algorithm. Presented Features are Norm-1 & energy CTA extracted from different sections of bi-frequency plane. This layout illustrate good characteristic in database. In addition, this manuscript illustrates a novel framework for generating textural and spatial information, and higher retrieval percentages. The textural features extracted with proposed CTA utilizing first & second moments among the image tiles is so effective in data processing. Spatial information is extracted utilizing decent field matrix (DFM). After that, moments are computed from DFM to get spatial features. The composition of the textural features and conjunction with the spatial information leads to a fantastic features matrix for retrieval. The experimental results on database guaranty the method efficiency on all classes of database with more than 10000 image. For measuring the distance of features a simple matching system based on Minkowski & Canberra distances is introduced. The results are compared with previous scholars and retrieval percentage is increased more than 10% in comparison with previous systems.