Influence of Boundary Extraction Operators on Boundary Shape Classification Algorithms Based on Complex Networks'
Sheng-Qiao Bao,Huibin Xu,Hao Wang,Jie Liu
DOI: https://doi.org/10.1109/cac51589.2020.9326957
2020-01-01
Abstract:As an important visual attribute for characterizing image objects, shape is always be investigated by researchers in pattern recognition and image analysis fields. It provides the most relevant information about objects for identification and classification tasks. By using the classical first-order and second-order operator algorithms, this brief paper extracts the shape boundaries of contour objects in several image sets. Then, generating explanatory variables based on the newly proposed complex network analysis method of the associated dynamic evolution boundaries network was carried out. Differences in the construction of high-dimensional feature interpretation vectors of boundary shapes based on network analysis are investigated. Our research shows that: the shape contour fingerprint vectors constructed by contour network construction algorithms based on different boundary extraction algorithms, can effectively represent the internal geometric figures of contour graphics features. The extracted digital feature vectors can be successfully used for graph classification tasks on the Etu10Silhouette database and Kimia216 database. There are some differences in CPU costs and classification accuracy. For the Etu10Silhouette database, it is recommended to use Sobel operator to extract the contours of the image shape, then use the algorithm mentioned in this paper to extract the fingerprint vector, combined with the KNN classification algorithm, one can get the best classification accuracy with almost the least time overhead. For the Kimia216 database, it is recommended to use the Laplacian operator to extract the contours of the image shape, with the boundary network fingerprint vector extraction algorithm, and combined with the LDA classification algorithm, one can obtain the best classification accuracy with almost the least time overhead. The comparative study on the classification algorithm based on deep learning method based on CNN also showed that, based on the graphic boundary shape analysis and classification method described in this paper, the classification method is simple and easy to be implemented. Our method has potential application prospects in industry.