Study In Nature Image Retrieval Based On Image Semantic

With the development of network and multimedia, the number of digital image keeps increasing. Content-base image retrieval (CBIR) arises at the historic moment. But in the practical application CBIR system discovered that users must provide the inquiry image, Then extract the low features form the inquiry image and find the closest image in the image database through compare their low features with the inquiry image's. But when users have not inquiry image, only have some fuzzier concept about inquiry image, then CBIR appears helplessly. In recent years many scholars propose image semantic to the image retrieval system. The solution of "semantic gap" which existence between the low-features and the high-level semantic features has become the key in problems of the semantic image retrieval.Many scholars use mapping methods which make a image belongs to a kind of semantic image.In the natural scenery image which contains rich semantic, and their classification are not very obvious. For example the image belongs to several kind of semantic image which includes mountain, water and blue sky. This paper proposed use the low feature in the nature image automatically obtains the high level semantics. And complete multi-semantics classification, finally realizes effective retrieval in the natural scenery images. The paper main works include: (1) This paper presents repeated using the optimal threshold for a roughly extract the largest target area of the color image, which base on the color of target area comparatively singleness in the natural scenery images. Then improved K_means clustering algorithm used to improve the accuracy of the segmentation from the target area. Experimental results show that this method can effectively extract color image of the object. It is also a certain degree of robustness to the noise image. (2) Base on the largest target areas, color and shape features are extracted. Then the fuzzy nerve network is used to map low-features and the high-level semantic features. So this method has realized information of image's attribution effective transmission and automatic extract high semantic from the image. And the experiment proved which has obtained the high accuracy. The accurate retrieval rate approaches 90% and the recall rate also achieved 75% in some class image of the nature image database. The experimental result has proven this method has validity and advantage in the natural image retrieval.