Research On Multiscale Shape Description And Retrieval Algorithm

As the low-level feature of image,shape description and retrieval are widely used in image processing and computer vision.Under a single scale,we cannot effectively describe and retrieve the shape.Specifically,if the scale value is set too low,although it can capture the local details of the shape,there will be a lot of redundant information in the extracted shape features.If the scale value is set too high,it can obtain the overall features of the shape,but the noise or the details of the shape will lead to classification confusion.Therefore,this paper studies the multiscale shape description and retrieval technology,and proposes three effective multiscale algorithms by reasonably constructing the scale space of shape.The main contributions and innovations of our work are as follows:A shape matching algorithm based on rectangularized curvature scale-space maps is proposed.In order to solve the problem of oversimplification of the existing curvature scale space technology in the shape description process,we use rectangularized box to restrict the arch contour of the curvature scale-space maps,so that we can capture and preserve the width and height features of the arch contour simultaneously.Then we apply circular cross-correlation operation for shape matching.It evaluates the similarity between shapes,and realizes fast shape retrieval from 8 steps to 1 step.Experimental results show that the proposed method has obvious advantages over the existing algorithm in retrieval efficiency and computational complexity.A shape description and retrieval algorithm in fused scale space is proposed.On the basis of analyzing the advantages of morphological scale space and Gaussian scale space,we combine morphological parameters and Gaussian parameters to form a fused scale s-pace.And we describe the shape based on height function descriptors across multiple scales,which can deal with the strong noise,intra class transformation and irregular deformation of the shape simutaneously.Then,we propose a multiscale fusion strategy in shape matching to get the final retrieval results.Experimental results show that the retrieval rate,robust-ness and computational efficiency of our method are better than the current shape retrieval algorithms.A multi-scale shape description and retrieval algorithm based on azimuth is proposed The existing height function descriptors calculate the height value according to the pixel position,which leads to repeated extraction of the shape features in the same azimuth.In order to reduce the redundant information in shape description,a strategy of extracting shape features directly according to azimuth is proposed.At the same time,the multiscale shape description based on azimuth is formed by combining the advantages of fused scale space in linear and non-linear shape transformation.In addition,two global parameters,center rate and eccentricity,are introduced to get the retrieval results in the fused scale space.Experimental results show that the algorithm has good performance.