Research On Content Based Shape Retrieval

Shape description and matching is an active topic in computer vision and pattern recognition which has been widely applied in many industrial applications and scientific research fields such as image retrieval,medical imaging,and product identification.The main idea is to use a computer to simulate the physiological process of the human eye's retina to perceive objects,to extract the binary image-related features from the whole or part of the object,and to use some classical pattern recognition algorithms to convert it into an easy expression form.The various images in the data set are matched,and finally the images that are close to the retrieved image are returned in order.The main work of this paper is mainly based on the shape description and analysis methods.Based on the existing shape retrieval algorithms,the improved algorithms are proposed from retrieval accuracy and shape matching speed,and classical methodologies are integrated with the proposed algorithms.The main contents and contributions of this article are as follows:?1?First,through the contour-based and region-based two categories,we introduce some classic shape-compact description and analysis methods from shape retrieval to the present.We describe in detail the essential ideas and implementation process of these algorithms,and compare their advantages and disadvantages,and analyze the possible improvements.?2?Second,for the CPDH descriptors and IDSC descriptors have high complexity and shape retrieval speed is slow.The EMD-₁ algorithm is used to calculate the similarity between two-dimensional histograms.The essence of the EMD-₁ algorithm is based on the EMD?Earth Mover's Distance?integration into the ₁ norm,calculate the ground distance matrix in flow way.The problem of optimal matching between two-dimensional histograms corresponding to shape in IDSC can also be solved by using EMD-₁ similarity measure algorithm instead of dynamic programming algorithm.Experimental results show that algorithm not only can effectively improve the retrieval and matching speed of shapes.But also maintains good scaling and translation invariance.?3?Third,CPDH is a good two-dimensional shape descriptor,but its poor retrieval performance in large data sets.Aiming at this shortcoming,a shape matching algorithm of cooperative delivery mechanism is proposed.The basic idea of the algorithm is to build a label transfer framework called Co-transduction based on Labe propagation?LP?in semi-supervised learning.By running two different search algorithms on similarity matrix,and a tag information the query shape is obtained,in the two similarity matrix for iterative retrieval and label propagation,and finally return to its most similar shape.Under the large-shape dataset,the retrieval performance of the improved algorithm is obviously better than the CPDH algorithm.?4?Fourth,for the measurement of similarity between CPDH a pyramid matching algorithm based on shape contour features is proposed.Different from other traditional histogram measurement algorithms,the pyramid matching algorithm divides the outline of the shape into several blocks,and assigns corresponding weights to each block,and then statistically calculates the weighted sum of feature calculations in the block.Finally,the similarity between two-dimensional histograms is calculated.The complexity of this method is low,By a large number of experiments,it is shown that this approach can achieve better retrieval accuracy and recall rates than that of an object area based only.?5?Fifth,CPDH is a shape descriptor based on statistical ideas,but it lacks the necessary discriminatray information.Linear Discriminant Analysis?LDA?and Kernel Discriminant Analysis?KDA?methods are introduced on the basis of the CPDH,and the DCPDH algorithm is proposed.The experimental results show that the recognition accuracy of DCPDH in large data sets is obviously better than CPDH.The retrieval speed of DCPDH has also been significantly improved,and it maintains good robustness.