The Research On Contour And Skeleton Sequence Coding For Two-Dimensional Shape

2D object recognition is a classic task in the field of computer vision and artificial intelligence,which is widely used in various applications such as image understanding,target recognition and scene analysis.The basic problem of object recognition is the representation and description of objects,which is usually based on the color,texture and shape of objects.Among them,shape may be the most intuitive and important description of objects.Through the shape information,humans can easily recognize different objects and their categories.Due to the above advantages of shape representation,a topic of particular interest is to represent and recognize objects by their shapes.Recent advances in two-dimensional shape representation and recognition includes methods based on deep neural networks and methods based on bioinformatics.However,the deep neural network architecture is mostly designed for the field of two-dimensional image analysis,and generally requires large-scale labeled samples.The size of the public two-dimensional shape data set is often rather small,which restricts the deep neural network in the two-dimensional shape field.The basic idea of the two-dimensional shape representation based on bioinformatics is to convert the outline of the two-dimensional shape into a biological information sequence,and use the standard biological information sequence analysis tools to match and recognize the two-dimensional shapes.However,there are still some problems in the current shape recognition based on bioinformatics.Firstly,the shape is encoded by the outline of the shape,and the encoding sometimes generates redundancy,and the recognition accuracy is not high in the experiment.Secondly,the existing 2D shape biological information encoding technology does not fully consider how to encode the shape sequences with more genetic information.Thirdly,the dual-sequence comparison tool is frequently used in the matching stage,and the matching efficiency is not high.To address the above issues,this paper proposes a novel two-dimensional shape recognition method based on bioinformatics.This method encodes the shape into biological information sequence by combining the advantages of the shape outline and the shape skeleton,and uses biological gene sequence comparison tools to analyze and identify it.Specifically,in the process of encoding the shape,this paper uses the skeleton to represent the strand branch of the shape,proposes a joint representation of the shape contour and the skeleton,and separately encodes the contour and the skeleton in different types to reduce coding redundancy.Then,this paper discusses the impact of different encoding methods on shape recognition results in conjunction with existing biological information analysis tools.Finally,this paper combines global descriptors such as shape area,eccentricity or deep feature descriptors to further improve the accuracy of the proposed shape representation and recognition methods.In order to evaluate the performance of the proposed method,this paper conduct shape recognition experiments on four public shape data sets including MPEG-7,Animal,ETH-80,and Swedish leaf,and compares the recognition accuracy with various existing shape recognition methods.Moreover,this paper further analyzes the influence of different shape coding schemes and different global descriptors on shape matching accuracy.The experimental results in the four databases demonstrate that the proposed method achieves a higher recognition accuracy rate,and shows the effectiveness of our sequence coding strategy by combining the contour and skeleton as a joint representation of the shape.