Research Of Image Classification Algorithm Based On Improved Local Binary Patterns

In recent years, with the improvement of people’s life, image acquisition and processing equipment is becoming commonplace. We can get image from the camera, mobile phone and other equipment. With the rapid development of the Internet, in order to effectively manage image information and reduce the labor cost, computer should be able to automatic distinguish image information, so as to realize the classification management.Image classification is a research field that across a range of scientific areas, it plays an important role in many disciplines such as pattern recognition, machine vision and so on. Image classification technique extracts image feature through the feature extraction method, then classifies the image through the feature. Image classification mainly includes two techniques: feature extraction and classification algorithm. The image features are divided into low-level feature and high-level semantic feature, now the low-level feature is more mature. The low-level features include color feature, shape feature, texture feature and spatial relationship feature, each feature can be used alone, and also can have a mixed use to improve the classification effect. The purpose of classification algorithm is classifying the image through the features. Recently, artificial neural network, support vector machine and some other classification algorithms have become the focus in the research of computer vision.LBP(Local Binary Patterns) is an excellent feature extraction algorithm. Because of its simple algorithm and low computational complexity, it is widely used in face recognition, scene classification, medical image processing. This paper mainly studies the LBP algorithm and obtain more comprehensive feature to improve the performance of image classification system. The main contents are as follows:(1) In this paper, the research on image classification field is reviewed. Then make a brief description of the current research of image classification and introduce the application of image classification technology in scientific research, military and other fields.(2) Introduce the relevant technology of feature extraction, especially the extraction of shape feature, color feature and texture feature. Then analyze the advantages and disadvantages of various algorithms. We have completed an overview on the classification algorithm and introduce the research status of classification algorithm.(3) This paper expounds the idea of the Local Binary Patterns, introduces the research status, improved algorithms and related applications. LBP has a good performance, but the traditional LBP only extracts feature in gray areas and does not consider the spatial structure relationship in the image. To solve these problems, this paper puts forward the improved algorithms. First use the channels of the RGB and HSV color space to fuse the color feature and texture feature, and propose local binary patterns based on color space fusion. Then extract the spatial structure relationship that may exist in the image through two adjacent regions, and propose adjacent local binary patterns based on color space fusion. The methods in this paper enhance the comprehensive of feature and improve the classification effect.(4) In this paper, the improved LBP algorithms are applied to build an image classification framework. In the experiment, we use SVM to classify the images. The datasets are Corel-1k dataset and MIT Vision Texture dataset. In order to test the performance, we will compare our methods with CSLBP, ULBP and CLBP. Experiments show that the methods of this paper can improve the accuracy of image classification.