Research On The Key Technology Of Traditional Culture Patterns Segmentation

Traditional cultural patterns are widely found in cultural relics and historic sites,cave murals,art works,ethnic buildings,ethnic costumes and home decoration,etc.They are important carriers of Chinese fine traditional culture.With more and more attention paid to traditional culture,the integration of technology and culture has become a research hotspot.How to use image segmentation technology to segment the sub-patterns with different cultural connotations form the traditional cultural patterns will provide technical support for the subsequent application of traditional cultural patterns in the fields of cultural gene interpretation,analysis of cultural space-time evolution,protection of cultural inheritance and design of cultural value-added products.Image segmentation,as one of the fundamental and challenging tasks in the field of computer vision,has been widely concerned by researchers.With the development of technology,the segmentation task has been put forward higher requirements.In recent years,deep learning-based segmentation algorithms have achieved remarkable segmentation results on natural images,but the segmentation effect on traditional cultural patterns is not ideal,because it is difficult to obtain enough training samples for traditional cultural patterns.In order to achieve the following specific objectives in the segmentation of traditional cultural patterns,that is,the fast segmentation that preserves the edge information of the sub-patterns,the global adaptive segmentation of complex content patterns,and segmentation that satisfies the cultural implication integrity of different sub-patterns,this dissertation carries out specific research work on the existing basis.The main contributions and innovations of our dissertation are shown as follows.(1)Research on fast segmentation method of traditional cultural patterns based on superpixel multi-feature fusion.In order to rapidly segment the high resolution traditional cultural patterns containing fuzzy edge sub-patterns and to maintain the authenticity of the sub-patterns edge,our dissertation presents a fast segmentation method of traditional cultural patterns based on superpixel multi-feature fusion.The proposed algorithm improves the ability to capture fuzzy boundary information of sub-patterns,and ensures the authenticity of sub-patterns edges by designing multi-feature fusion quantization expression algorithm based on superpixel histogram of oriented gradient(HOG)feature.The proposed algorithm can effectively improve the segmentation efficiency of high-resolution traditional cultural patterns by transforming the images to super-pixel level graph model.(2)Research on adaptive segmentation method of traditional cultural patterns based on the SLEGM.In order to achieve the effective features quantitative expression of the traditional cultural patterns with complex contents,and to meet the requirements of adaptive segmentation of the sub-patterns considering the global information,this paper proposes an adaptive segmentation method of traditional cultural patterns based on the superpixel Log-Euclidean Gaussian metric(SLEGM).By designing a feature quantization expression algorithm based on the SLEGM,the feature quantitative expression in Euclidean space is transformed into Manifold,so a more effective feature quantitative expression for segmenting to traditional cultural patterns is obtained.Because the space of Gaussians sample covariance matrix distribution is not a linear space but a Riemannian manifold,we map this manifold via matrix logarithm into a linear space,which enables us to handle Gaussians with Euclidean operations.In the segmenting stage,an improved spectral clustering segmentation algorithm considering the global information of the image is proposed,which realizes the adaptive decision of the number of clustering centers of the superpixel sparse adjacency matrix based on the SLEGM features quantitative representation,and reduces the over-segmentation to a certain extent.(3)Research on the multi-scale segmentation method with structural consciousness for traditional cultural patterns.In order to segment as many sub-patterns with different cultural implication as possible in traditional cultural patterns,so as to meet the requirements of the cultural implication integrity segmentation of traditional cultural patterns,this dissertation proposes a multi-scale segmentation method for traditional cultural patterns,and the segmented regions are guaranteed to be consistent at different resolutions.Our dissertation proposes a contraction and fusion iterative optimization segmentation method,which realizes the region division of traditional cultural patterns by maintaining the target consistency,and finally completes the construction of the hierarchical structure of segmented patterns based on the shape space.By proposing the salient target selection based on causality principle and the target hierarchy extraction based on morphology,the branch adjustment of the hierarchical structure is realized.By designing the hierarchical adjustment and scaling optimization technology based on the scale marker,the adjustment operation of different levels of the hierarchical structure is realized.On the basis of the above regional division and hierarchical adjustment,the multi-scale hierarchical segmentation with structural consciousness of the sub-patterns representing different cultural implication in the traditional cultural patterns is realized to a certain extent.Our dissertation only focuses on the specific target of traditional cultural pattern segmentation,and studies the key technologies encountered in the task of traditional cultural pattern segmentation,and puts forward the corresponding algorithm model.By comparing the proposed algorithms with the state-of-the-art algorithms of the same category on Our Chinese Cultural Pattern Database(Patternnet)and the Berkeley Segmentation Data Set(BSDS500)benchmark,the experimental results verify the effectiveness of the proposed algorithms.The proposed algorithms achieve a satisfactory segmentation effect on the traditional cultural patterns,and the segmentation results are more in line with the actual needs and close to human subjective cognition.