Improved Directional Local Extrema Patterns as a feature vector for CBIR

In order to get the local structures of an image, the Local Binary Patterns and its variants are used. These are based on obtaining the difference in intensity values of the target pixel and its neighbors and assigning some value to the center pixel. However, the directions are not considered in these patterns. The Directional Extrema Patterns are used to encode the relationship between the reference pixel and its neighbors by computing the edge information in four directions. The complexity is high when the DLEP is used as a feature vector whose size is n × n. The proposed work aims at developing a feature vector for Content Based Image Retrieval system. Further, the search for similarity can be made simple by discarding many images at every level by implementing the search tree approach due to which the retrieval speed increases which is a primary concern in image retrieval.