Image Retrieval Method Of Multi-feature Measurement Fusion Based On Improved Hausdorff Distance

The image has become the most efficient way of information storage,transmission and expression,because of its outstanding advantages of containing a large amount of information.A variety of mobile terminal equipment and digital cameras and other image devices have been universal and the number of images of resources was massive.With the rapid development of computers and the internet,image application as represented by network was explosive growth.Image retrieval technology is an effective way for searching the desired image in massive digital images,but its development is far behind the development and application of digital image itself.Therefore,the research of image retrieval method has important theoretical and practical value.Image retrieval technique can be summarized as Text-based image retrieval technique and content-based image retrieval technique.In contrast,the latter is more suitable for large-scale digital image retrieval owing to its automatically extracting image features and understanding the image content,thus its development is the inevitable trend.The current Content-based image retrieval research is focused on the underlying characteristics.The two keys are respectively selection and its extraction of image feature,and similarity distance and its measurement.Based on these two keys,the purpose of this paper is to improve image retrieval performance.This paper mainly studies image retrieval methods and its related problems from the perspective of similarity measure.Currently,distance measurement method is usually used,in which the Hausdorff distance is superior to the commonly used Euclidean distance,and is being extensively applied.For the image retrieval method based on content,this paper uses the Hausdorff distance to make the similarity metric,and improve it to achieve better image similarity measurement.Thus an improved Hausdorff distance measurement method was proposed,and cost function was constructed as Norm distance to adjust the original distance value.The proposed method can reflect the image of the overall similarity and can reduce the abnormal point,occlusion,scene changes and complex background interference.Then through the combination of single feature and measurment,i.e.color histogram and texture gray level covariance,a variety of typical distance measurement methods were used to verify the improved Hausdorff distance.In addition,for the image feature selection and extraction link,through the multi-feature fusion,the more accurate image retrieval results can be got.For improving the accuracy of image retrieval,this paper focuses on the multi-feature fusion image retrieval method based on DS evidence theory.We constructed and implemented a method of multi-feature measurement with equal weight addition image retrieval based on improved Hausdorff Distance.By this method,for improving the performance of image retrieval,the improved Hausdorff distance is used to measure the similarities of multiple features,and these similarities added with equal weight were as a measurement result of similarity.This paper proposes a multi-feature measurement image retrieval method based on DS fusion.In this method,an improved hausdorff distance was used to measure similarities,and DS theory was used to fuse multi-features as similarity measurement result,which can significantly improve the accuracy of image retrieval.For verifing this fusion method,two feature measurement fusion of color histogram by BOW model and texture gray scale covariance was used.Furthermore,for large-scale image datasets,Scalable Vocabulary Tree(SVT)image retrieval method was researched.Scalable Vocabulary Tree was constructed and by this image retrieval method,large-scale image retrieval can be realized.We proposed a fusion image retrieval method of multi-feature measurement based on SVT.Improved Hausdorff distance was used to many features,and improved DS combination rule was used to many features as similarity measure results,which can improve image retrieval accuracy.Based on dense patch SIFT descriptors and dense patch DAISY descriptors,we used histogram features image signature,kernel density image signature,Euclidean distance,CFHD distance to realize SVT image retrieval method.This paper presents an image retrieval comparison experiment with the existing mainstream image retrieval method,and verifies the proposed method.Research indicates that,the proposed improved Hausdorff distance has strong anti-interference and has better image retrieval performance;precision of SVT applying for large-scale image retrieval has been improved;multi-feature image retrieval method of DS fusion can significantly improve image retrieval performance.The proposed method provides an effective way and technology base for improving anti-interference capacity and image retrieval performance,and has promotion effect for image retrieval development and application.