Research On Texture Features Of Texture Image

Image recognition,classification and scene classification of remote sensing images are inseparable from the description and extraction of image features.In the field of image feature research,texture is one of the most frequently used features of image.Texture mainly describes the structural features of the image surface,and reflects the brightness transformation and spatial distribution of information.However,although the human visual system can easily and accurately identify and describe the texture,it is difficult to extract the texture features for texture classification and different application areas.In this paper,we study the method of texture periodicity and directionality of image texture based on the theory of variogram.The variogram can reflect the texture and randomness of texture image.This paper calculates texture from random direction.But the main research from the row and column direction,draw the variogram,to achieve the texture image row and column periodicity and directionality analysis.The experimental results show that the proposed method can effectively analyze the periodicity and direction of the texture image,compared with the autocorrelation function and the distance matching function,and the texture period and image texture periodicity error is very small.Secondly,according to the problem of wallpaper period measurement,this paper found that the algorithm can accurately detect the cycle of wallpaper and printed fabric,and the segmentation unit of wallpaper is effectively divided,which is the key to realize the production automation,Significance.Then,for the problem of image classification,regularity is an important feature of texture images.Based on the variogram,the method of regularity calculation is proposed,and the images are classified according to rule-approximate rule-random texture model and the thickness of textures.The feature vectors and image classification models of periodic regularity feature are constructed by using the quadratic derivative and the threshold division to extract significant peaks and valleys.With the Brodatz image library The experimental results show that the regularity can effectively distinguish the regular image,the approximate regular image and the random image,which is in accordance with the visual perception,and has been compared with the existing research results and has better accuracy.Finally,this paper makes a more accurate classification of features from several aspects,such as feature extraction,wallpaper period determination,texture classificationand image retrieval technology.