Discrete Sine Transform Sectorization for Feature Vector Generation in CBIR

We have introduced a novel idea of sectorization of DST transformed components. In this paper we have proposed two different approaches along with augmentation of mean of zero and highest row components of row transformed values in row wise DST transformed image and mean of zero- and highest column components of Column transformed values in column wise DST transformed image for feature vector generation. The sectorization is performed on even-odd plane. We have introduced two new performance evaluation parameters i.e. LIRS and LSRR apart from precision and Recall, the well-known traditional methods. Two similarity measures such as sum of absolute difference and Euclidean distance are used and results are compared. The cross over point performance of overall average of precision and recall for both approaches on different sector sizes are compared. The DST transform sectorization is experimented on even-odd row and column components of transformed image with augmentation and without augmentation for the color images. The algorithm proposed here is worked over database of 1055 images spread over 12 different classes. Overall Average precision and recall is calculated for the performance evaluation and comparison of 4, 8, 12 & 16 DST sectors. The use of Absolute difference as similarity measure always gives lesser computational complexity and better performance.