ABSTRACT

This paper introduces a novel idea of Binary Pattern observation of column wise and Row wise Walsh transformed color images for feature vector generation. The density distribution of Sal, Cal components of Binary Pattern and its mean values are considered as two different approaches for the same. The proposed method experimented with and without augmentation of average of zeroth Cal and average of last Sal components in the Feature vector. This paper discusses use of Euclidian distance and sum of absolute difference as similarity measures in each of the approaches as mentioned above to check the retrieval performance. The work is experimented on image database of size 1055 images containing 12 classes. Two new parameters of performance measuring parameters i.e. LIRS and LSRR are introduced apart from precision and recall which are very general.