Study And Comparison On Image Texture Featur

An image can be represented by color features, texture features, shape features or spatial features. Compared with other image features, texture features seem to take into account the macroscopic properties and microstructure of an image. So, the texture analysis becomes an important means of image analysis. Presently, texture features have been successfully applied in many industrial areas, which brings the important practical significance. There is no single definition and precise standards for texture, and so it becomes more complex and challenging to analyze texture.This paper describes four major categories of texture representation separately in detail. They are statistics methods, modeling methods, structure methods, and signal processing methods. Their advantages and disadvantages are compared in qualitative way.By qualitative comparison, it is found that the complexity of traditional texture features limits its wider application. Local Binary Pattern (LBP) is originally used for manual texture classification. Because of its simple complexity, gray-scale invariance and rotation invariance, it is used in many applications, such as in age estimation, background modeling, face recognition, sex determination and so on. Though the original LBP operator performs better in many applications, it has redundant information. In this paper, in order to remove redundancy, a simplified LBP operator is proposed based on information entropy theory. Meanwhile, according to the principle of run-length coding, texture feature that equal LBP run-length statistics based on zigzag structure is proposed.In order to compare the performance among S-LBP and other texture features in quantitative way, the license plate character recognition system and crowd estimation system which both belong to intelligent video surveillance system are designed to evaluate the texture features performance. Experiments show that the LBP and S-LBP operators have a distinct advantage over others either in recognition accuracy or in recognition speed.