Font Size: a A A

Bilateral Local Binary Patterns For Rotation Invariant Texture Retrieval

Posted on:2016-08-06Degree:MasterType:Thesis
Country:ChinaCandidate:M G ZhouFull Text:PDF
GTID:2308330461467409Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
Local Binary Pattern (LBP) is a simple algorithm proposed in recent years, which has strong ability for identification of rotation invariant texture image. As important structural features of the image, Phase can describe the important information of the image. Based on the research of the original LBP operator, this paper proposed a new operator called bilateral local binary pattern that can efficiently extract features of rotation invariant texture image based on module and phase information.The main works of this paper are as follows.Firstly, we introduce the local binary pattern operator in detail. LBP is a simple algorithm with low computational complexity, strong ability and wide usefulness. However, LBP use a single mapping for all problems, which leads to poor description of the image features, such as angle, position, structure. We propose a new feature to represent texture image combining phase information which reflects the position and the amplitude values which reflects the strength of varies.Secondly, we propose the method of acquiring the phase information of image. We draw on the idea of the complex wavelet transform based on analytic signal and utilize the analytic signal based on projecting filter to process the two-dimensional images. And then we can obtain the complex-valued image and phase information. The obtained phase information retains the feature information of image better and it is shift invariant.Thirdly, we propose texture spectrum BLBP. We draw on the idea of combining phase information with LBP for rotation invariant texture image retrieval. We filter the two-dimensional image with projecting filter to obtain the phase information, and then are transformed by the LBP operator to create the histogram of phase information. The original image is transformed by LBP directly to obtain the histogram of module information. We make the two histograms together to represent the texture image feature. BLBP operator retains the advantages of original LBP operator, such as simple algorithm, low computational complexity and so on. BLBP adds the phase information of the image without changing the feature dimension, so that the BLBP can describe the image features well and it is robust for image retrieval.We present the texture retrieval experiments according to the proposed BLBP operator. The results show that BLBP not only keeps the advantages of simple algorithm, low computational complexity of the original LBP operator, but also achieves a higher retrieval rate.
Keywords/Search Tags:LBP, Projection, Analytic signal, Phase information, Texture retrieval
PDF Full Text Request
Related items