Scene Recognition Technology Research

Scene recognition is an important branch of image processing. MIT scene understanding conference in 2006 made a clear definition of the concept of the scene and scene understanding, and pointed out that the scene recognition had became a new promising research direction.For the demand of the process in automatic extraction and tracking objects with different scenario processing in image processing, this chapter presents a description of scene recognition based on the image complexity describing. And in view of the disadvantages of scene recognition simplicity using the image complexity, we proposed an algorithm based on the image complexity and the SVM classifier, in this paper.This paper mainly completed the following several aspects:At first, we have researched the background and significance of the scene recognition technology research in this paper, and summarized the technical difficulties that the scene recognition faced by. Then we study the selection and extraction of the image feature deeply, conclude the applicability of different image feature descriptors and summarize the advantages and disadvantages of them.In this paper, we put forward a description method of image complexity based on the underlying characteristics and use it for scene recognition. We also put forward a gradient descriptor——Gray gradient matrix due to the existing descriptor is not enough to describe the image complexity fully.We need to the weighted average image characteristics in the process of calculating the image complexity. In the process, we have lost the independence of the characteristic. In view of this point, we proposed an algorithm based on the image complexity and the SVM classifier, in this paper.We describe the image complexity by the underlying features, and recognition the scene in combination with the SVM classifier to guide the target extraction and tracking in different scenarios.