A Feature Extraction Method Based On Image Texture And SIFT Algorithm

Feature extraction is an essential part of digital image processing.As the basis of image classification,it has always demonstrated key value in practical applications.Texture is a major component of image features,with global patterns consisting of local sequential attributes describing visual properties with homogeneous phenomena and is widely used in areas such as face recognition,medical detection and remote sensing monitoring.With the rapid development of artificial intelligence,research on textured images is challenged by increasing demand.Among them,global feature methods are inefficient in recognition when images have similar distribution;local feature methods suffer from low operational efficiency,feature points dominated by areas of interest,and wide variation in number of dissimilar structures.Current methods just test the validity through single message,thus leading to low accuracy in analysis of complex patterns,which fails to meet the practical requirements.Therefore,in order to improve classification performance,this paper proposes a feature extraction method based on texture and SIFT algorithm,which fully combines good properties of SIFT key points,aiming at constructing descriptors with advantages of high discrimination,stability and robustness for the recognition of wild texture patterns.Specifically,starting with SIFT algorithm,on the one hand,using key point information to set an adaptive window,extracting multiple cell blocks by window transformation to make local features detailed and differentiated;on the other hand,texture analysis is carried out for local areas,combining co-occurrence matrix,with the grey scale distribution and gradient structure of statistical pixels to generate texture descriptors.Finally,the BoVW model is trained to vectorize the images by Mini Batch K-Means cluster coding to resolve the inconsistency between local and global feature dimensionality.During the process,the parameters of the model are adjusted in conjunction with the feedback from the indicators.The experimental results on DTD show that the improved feature extraction method makes full use of the stability and invariance of the key points of the original algorithm.It can effectively overcome the shortcomings of efficiency and feature expression capability.Furthermore,there is a significant improvement in the classification accuracy for complex texture patterns in real natural scenes.