Using LBP to Improve PCNN Performance for Texture Image Retrieval

Abstract

Local binary pattern (LBP) spectrum is a powerful feature for texture image, which is invariant to local illumination changes. Pulse coupled neural network (PCNN) is a biologically inspired algorithm, which is well suited for image processing and can generate rotation, scale, translation invariant image signature. This paper proposed an image retrieval tool named LBP-PCNN which combined the advantages of LBP and PCNN. First, images are mapped into local structural domain with rotation invariant LBP. Then, the simplified PCNN was adapted to extract the image signature in structural domain. At last, texture image retrieval was completed by measuring the signature similarity of input sample texture image and texture database. Experimental results on Brodatz texture gallery show that LBP-PCNN overbears LBP and PCNN in texture image retrieval and has potential applications.