| In recent years, with the rapid development of mobile Internet technology and the rapid popularization of digital products, the number of digital images increase in a geometric exponential growth, and how these images efficiently store and quickly retrieve is becoming a hot research. However, a problem existing retrieval techniques always need to face is the "semantic gap" phenomenon, namely computer graphics technology to understand the difference between high-level semantic and the underlying features issues. From the beginning of reason produces semantic gap we conduct a analysis in image retrieval technology research based on content. This article covers three aspects: ① significant improvement in image ROI(region of interest) extraction method ② using SIFT operator for local feature extraction ③ use local features for image retrieval.Visual attention once is the concept of neurobiology, which is external objects of human point and activities focused on the individual at the time of the heart; in short, which is get into the human eye at first when observing an image. Existing research shows that in the beginning of the observed image, the portion of image the first to get the attention of the human eye, which is often ROI or part of the area Through continuous research on biological vision theory, scholars have proposed visual attention model, which can obtain the region of interest in the image, this area can be expressed saliency features of the image. Since the background image retrieval secondary content occupy a lot of space, if we can eliminate this background, the main part of the image to use the feature match is bound to improve efficiency.This paper analyzes the research status situation based on the general method of content-based image retrieval and visual attention model, followed by studies of saliency extraction based on Visual attention model algorithm. Finally, SIFT algorithm is to be used to extract the image salient features of parts of area, which is to be used in image retrieval. The main contents are as follows:1) During the phase of map generation in Itti visual attention model, C-S operator is mainly used between the different levels of the Gaussian pyramid for the difference operation, the center is a high-resolution sample, is surrounded by a low-resolution sample. This will bring the following two questions:1) due to the surrounding area from the center of the sample was smoothed down, when they want to make the difference between the two enlarged to the same size, this will cause more significant figure fuzzy dark area.2) Gaussian pyramid decomposition into multiple levels of need, if the original image is blurry, then the resulting image after sampling almost impossible to identify. To solve these problems, this paper presents an improved method:For multi-level image, the central role of the surrounding operations only within each scale, rather than acting between the different levels, but by way of the window, in the present level of around inside the simulation center operations.2) Image retrieval methods need to extract image feature for feature matching, in order to accurately describe the main content of the image and can speed up the computation, this paper propose a method of only for salient region characteristics matching in retrieval, the use of the feature vector of salient region as an image matching similarity measure.3) SIFT operator is used for extracting the local feature point, which can be an object, regardless of the size of the object, the direction of rotation, and the light of fault tolerance, the noise, the higher the micro perspective change. This article will use SIFT algorithm in the extraction of the salient features of the region, and finally use SIFT feature vectors for matching feature.Experimental results show that the method in this paper on the recall and precision rate is higher than the global color histogram-based method, but also superior to multi-feature fusion method. |
PDF Full Text Request