Rearch On Image Local Feature And Its Application

The first important step is to extract local features from images in most tasks of computer vision and image processing, such as image classification, image retrieval, wide baseline matching, and so on. The performance of local features plays an important role in these applications. Consequently, the extraction method of local features is worth further studying. However, there are great difficulties when extracting local features from images due to many kinds of image transformations, such as scale, rotation, blur, background clutter, partial occlusion, viewpoint and illumination changes. Especially, the images normally contain complex background, noise interference and pose changes in real-world scenes. Therefore, great efforts are still needed because local features have important theoretical significance and practical value.Local feature design inspired by research results from local features is studied in this dissertation. The main contents and novelties of the paper are as follows:(1) The design of a novel descriptor called weighted center symmetric local ternary pattern.To better characterize the image local texture and achieve high distinctiveness, a novel descriptor called weighted center symmetric local ternary pattern (WCS-LTP), constructed by using the CS-LTP variance of the local region as an adaptive weight to adjust the contribution of the CS-LTP code in histogram calculation. Then, based on the proposed WCS-LTP descriptor, we introduce a new local WCS-LTP feature extraction approach. Finally, WCS-LTP feature based sparse coding spatial pyramid matching (ScSPM) representation classification is proposed for image classification. Extensive experimental results demonstrate that the proposed descriptor is distinctive for classificating images. And when the images contain complex complex background, significance noise interference and pose variation, the distinctiveness for classificating images is also demonstrated by experimental results. Furthermore, the proposed descriptor is robust to many image transformations including viewpoint changes, image blur and JPEG compression.(2) The construction of local feature combinating shape with texture information.Shape and texture information are critical to the accuracy of image classification systems. A new local feature called SIFT-WCS-LTP is proposed based on the WCS-LTP descriptor. Then SIFT-WCS-LTP feature based sparse coding spatial pyramid matching representation classification is proposed for image classification. Our proposed SIFT-WCS-LTP feature can not only capture the shape information of images, but also tend to extract more precise texture information. And it can better characterize the image information. Experimental results show that the effectiveness of proposed local feature. Furthermore, the proposed local feature also achieves distinctiveness when the images contain complex complex background, significance noise interference and pose variation.(3) Orthogonal symmetric local ternary pattern operator and description of image regions based on this operator.A key issue of making LTP a local image region descriptor based on a SIFT-like grid is to reduce its dimensionality while keeping discriminative power.A novel operator called orthogonal symmetric local ternary pattern (OS-LTP) is proposed, achieving robustness against noise interference and discriminative ability for describing texture structure. Then, based on the proposed OS-LTP operator, we introduce a new descriptor, named weighted orthogonal symmetric local ternary pattern (WOS-LTP). The WOS-LTP descriptor is constructed by using the OS-LTP variance of the local region as an adaptive weight to adjust the contribution of the OS-LTP code in histogram calculation. Experimental results show that the effectiveness of proposed local feature. Experimental results on various image transformations show its high robustness. And the distinctiveness for classificating images is demonstrated by experimental results. Furthermore, it is found to be computationally efficient.The research results not only have important significance for the development of local feature extraction but also play a beneficial role in image classification, image matching and wider areas that are relevant.