The Study On Library Retireval System Based On The Geometric Relationship With The Salient Features

This paper proposes a fast algorithm of effective books retrieval system, in theimage retrieval, based on the word with model algorithm is the traditional imageretrieval algorithm, there have been two major drawbacks, however, this algorithmfirst is local characteristics are visual words quantization error, affect the visualdiscrimination, and the visual words after quantitative are lack of geometric relations,thus affect the retrieval performance.Based on these two shortcomings, we proposed the retrieval algorithm based ongeometric relations between characteristic, the first to use significant informationcombined with SIFT features, get the local characteristics of significant areas, thenquantify the word, due to the significant characteristics filter the redundant features,thus improved the quantitative discrimination in the process of quantitative, thisextract significant area is based on the ITTI, uses a visual attention system, throughthe early visual system behavior and neural network, combines multi-scale imagecharacteristics and a single significant figure, with a dynamic neural network tochoose prominently, end up with significant area.We combine the significant regional with local characteristics, and based on thesignificant characteristics of spatial pyramid retrieval model are used to get the initialsearch results. This method through the image according to the average level ofdivided into blocks, and calculate the local histogram in each region. This spacepyramid retrieval model is a simple order word with retrieval model to calculate theeffective extension. Due to the traditional word joined the spatial relationships in themodel, we can effectively enhance the accuracy of retrieval, but the visual spacepyramid retrieval model is coded words of spatial relations, therefore it does notreflect spatial relationships between image features, and can’t completely reflect the retrieval system of the optimized, it will not significantly improve the retrievalperformance, so we put forward the method of geometric verification again,reorderingthe initial results of previous image geometric, optimize the retrieval results.Finally, we use a space based on the local characteristics of significantinformation retrieval model to get the first one hundred gold tower image geometricreordering, this article uses a fast effective geometric sorting method. Retrieves thesignificant characteristics of the image with the query image matching, will receivethe matching features of deposit in the position information of feature points in thematch list. And then only according to the location information in the list, we putforward of a method of location geometric similarity score, in the retrieval system forimage rotation, scale and displacement have invariance, three kinds of geometricverification algorithm was presented here, respectively is geometric similarity score,scale geometric similarity score, according to the analysis and comparison,locationgeometric similarity score more suitable for our retrieval system of geometricverification algorithm.To test the effectiveness of the proposed retrieval system, we according to thesignificant information’s contribution to the retrieval algorithm, and geometricverification algorithm’s contribution to the retrieval algorithm, we design the fivekinds of algorithm, and through comparing the accuracy and computation time of fivekinds of algorithm, the retrieval scheme we design greatly improve the retrievalaccuracy, the universality of retrieval methods in order to validate our design at thesame time, choose the non-book and many scholars commonly used database fortesting.