Texture Image Recognition Method Based On LBP Feature Extraction

With the rapid development of computers, image resources explosive growths,increasingly strong demand for image recognition, while the automation of image recognition requirements are also increasing. Pattern recognition refers to the pattern to be classified using computer technology to automatically(or adding a small amount of human intervention)correctly assigned to each model class technology, directly contributed to the development of the technology of computer applications, but also led to the artificial intelligence technology development. Image recognition has been one of the hotspots of pattern recognition which have been widely applied in reality.Image surface generally has some inherent characteristics, such as the gray statistics,spatial distribution information and picture structure information, these information together constitute an important feature of the image, the texture. Since the original image sample generally has high dimensions, pattern recognition method based on feature extraction extracted image feature information, achieved with a low-dimensional space to represent the sample, extracted sample has better separability, and relatively easy to design classifiers.Based on the background described above, application-specific texture image recognition,find a texture image recognition method can effectively portray the texture image features a very important practical significance. This paper introduces the concept and image pattern recognition methods, combined with the texture features, lists several traditional pattern recognition methods, focusing on the basic principles of LBP operator, characteristics,application, Experimental verification of the LBP operator on the texture image recognition by the direction of the feasibility and applicability, and proposed the use of cluster subdivision after the training model in image recognition to consume a small amount of time to increase the recognition rate direction.In view of the drawback of Local Binary Pattern(LBP) algorithm which is time consuming and has low texture image classification rate, a new improved LBP operator is proposed. First, Texture image equally divided into blocks, and LBP histogram obtained, then characteristics of the test sample collected according to the training model. Finally,multi-scale weighted characteristic vector distance used to classify texture images. Better classification results achieved in the NSF texture database, experiments showed that,compared with the texture image recognition method based on the LBP original, improved algorithms reduced dependence on the training sample texture, has a higher recognition rate as well as more stability.