Column wise DCT plane sectorization in CBIR

Content Based Image Retrieval (CBIR) is the application of computer vision techniques used to retrieve digital images from a large database. In this paper we have used the concept of sectorization of the plane formed form DCT transformed image. We have proposed an approach which involves augmentation of zero and highest row components of column-wise DCT transformed image for generating the feature vector. Precision-Recall crossover point, LIRS (Length of initial string of relevant images retrieved), LSRR (Length of string to recover all relevant images)and LSRI (Longest string of relevant images retrieved) are used to evaluate the performance of the proposed method. We have introduced a new performance evaluation parameter apart from the above mentioned parameters viz. and LSRI (Longest string of relevant images retrieved). Two similarity measures used to calculate performance evaluation parameters are: 1. Sum of Absolute Distance & 2. Euclidean Distance. The column-wise DCT transformed image is sectorized on the basis of even-odd column components of transformed image with and without augmentation of zero and highest row components. The proposed algorithm is applied to a database of thousand images spread over 10 varying classes. Performance is evaluated and compared for 4, 8, 12 and 16 DCT sectors. Keywords—CBIR, DCT, Precision-Recall, LIRS, LSRR, Absolute Difference, Euclidean Distance, Longest String of Relevant Retrieved Images.