基于bp神经网络的语义风景图像检索技术的研究-信号与信息处理专业论文.docxVIP

II II Abstract With the development of science and technology and Internet, the application of multimedia, lots of different information are being collected, transmitted and used quickly by people in the world. In the 1990s, Content-base image retrieval arises at the historic moment. It can estimate the comparability between images bases on image color, texture and shape. However, these image low-level features can only reflect objectively statistical features. And people can’t understand these low-level features immediately. Now people mostly base on image keywords to estimate the images which they want. The keywords may be blue sky, sunset or others. These keywords are also image high-level semantic features. It is obvious that image semantic retrieval can more meet the need of people .And it also has wider prospect. The solution of “semantic gap” which exists between low-feature and the high-level semantic features has become the key in problems of the semantic image retrieval. This paper establishes a multi-output BP neural network. This method extracts low-level features vector as the network’s input and the exceptions as its output. The system trains the network with improved BP arithmetic. When the training is over, this network can classify natural images. So it has established the mapping between the low-level features and high-level semantic features. The paper main works include:(1)Considering natural images have vivid color characteristics, by studying the RGB and HSV color space model, The paper presents eight natural main colors as color vector. And selects the appropriate texture and shape vector.(2)The method firstly segments the images and abstracts the low-level features from the segmented areas ,uses the BP neural network to establish the map between the low-level feature and semantic features .So this method fills up the “semantic gap”. In addition, the paper learns how to select the hidden layer nodes and the structure of training sample, so tha