Local mean differential excitation pattern for content based image retrieval

Abstract

Effective and facile local texture feature descriptors assume a significant role in many image classification and retrieval tasks. However, conventional feature descriptors are impaired to capture salient image features like local inherent structure, orientation and edge information of the image. A typical local binary pattern (LBP)-like feature descriptors elicits information based on the gray level difference from each locality and consequently its value immensely susceptible to noise. To overcome a few deficiencies of traditional methods, the proposed research work acquaint with a lucid, novel, yet robust texture feature descriptor called local mean differential excitation pattern (LMDeP) for efficient content based image retrieval. The main strategy of LMDeP is to elucidate differential excitation using the mean of points over each angular and radial neighbor points. This enables the LMDeP to fetch robust features by skimming the noise effect of enticing neighbors over each local patch. LMDeP performance compared to the LBP variants on Corel-1 K and Corel-5 K datasets establish the superiority of proposed method to its counterparts.