Improvement Of LBP Rotation Invariance And Its Application In Fabric Image Retrieval

As the target object inherent attribute,texture is the basic and important visual features of image,can effectively describe the image.Local Binary Patterns(LBP)is a local texture description operator,which is widely used in combination with statistical and structural methods,and is simple and reliable.However,when we use the LBP texture feature to make fabric image retrieval,we find that the general LBP texture feature is not ideal in the image library with rotation condition.The existing methods of improving the rotation-invariant of LBP can improve the retrieval recognition on the image with rotation condition,but the retrieval recognition rate on the image without rotation decline.This paper focus on the problem of image retrieval based on LBP texture feature is not related to image rotation,and it is aim at solving the problem that the image retrieval recognition rate caused by the direct correlation of LBP operator from fixed position is reduced,So,a method based on local average Direction of the LBP rotation invariance is proposed,which seeks a direction in the local domain that can rotate along with the image,with the adjacent pixel in the direction as the LBP operator encoding the starting position,so that the LBP operator has a rotation invariance while reducing the influence of the decrease in the retrieval recognition rate on the general image(without rotation).The main contents and innovations of this paper are as follows:(1)Research on LBP rotation invariant coding method based on Local Average Direction.First,on the basis of horizontal and vertical direction of the gray gradient,we add ± 45 degrees direction of the gray gradient to define the pixel point direction.The direction of the pixel represents the direction of rising(or descending)of the gray scale,and as the encoding start position will interfere with the LBP operator generation(the first fixed is 0 or 1),so a general local average direction and a more stable weighted local average direction are proposed,and confirm the local radius and weight parameters with the smaller rotation error on the experimental test.(2)Contrast test and result analysis of LBP rotation invariant coding method based on local average direction.Several LBP operators,which are commonly used,are coded to extract texture features and experimentally tested on different standard texture libraries.The results show that the ability to deal with rotating images of the LBP texture features are greatly improved by using the local average direction.Compared with other LBP operators with rotation invariance,the recognition rate of image retrieval without rotation is better and the applicability is stronger.(3)Application of LBP Rotation Invariant Coding Method Based on Locally Average Direction in Fabric Color Image Retrieval.Based on the improved rotation independent LBP texture feature and the color feature,a method of texture and color feature is proposed.The fabric image retrieval system is designed and developed,and the fabric image classification management model based on texture and color feature is established.But also we use the real fabric image on the system for different methods of the performance of the experimental test.