Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
A Novel Matting Algorithm based on Mahalanobis Distance and Fuzzy C-Means Clustering
1University of Science and Technology Beijing, Beijing, China.
2Beijing Cuiwei Primary School, Beijing, China /EADDRESS/3School of Mathematics and Physics, University of Science and Technology Beijing, Beijing, China.
4China Electronics Standardization Institution, Beijing, China /EADDRESS/
ABSTRACT
This paper describes a novel matting algorithm. First this algorithm calculates the Mahalanobis distance between each pixel in an image and a specimen in the background of the image. Then using fuzzy C-means clustering algorithm classifies the calculated Mahalanobis distances into two categories: the foreground and the background. Finally, using filling-hole technique improves the quality of the matting. This essay compares our algorithm with other Grow-Cut algorithms, Mahalanobis algorithm, KNN algorithm and Normalized Regression algorithm via experimental simulations. Experimental results show that the proposed algorithm can automatically segment the images with relative monotone color backgrounds in sound accuracy. Comparing with users' interaction methods, the novel algorithm is automatic, less time, and has lower complexity. For images with small random noises, segmentation results are also well looked. Also related to some specific problems, the algorithm get a better results too. In order to obtain quantitative evaluations to all the segmentation results, this essay put forward matting error to identify the results.
Keywords: Image matting, Mahalanobis distance, Fuzzy C-means clustering, Filling-holes.
Download PDF
1University of Science and Technology Beijing, Beijing, China.
2Beijing Cuiwei Primary School, Beijing, China /EADDRESS/3School of Mathematics and Physics, University of Science and Technology Beijing, Beijing, China.
4China Electronics Standardization Institution, Beijing, China /EADDRESS/
ABSTRACT
This paper describes a novel matting algorithm. First this algorithm calculates the Mahalanobis distance between each pixel in an image and a specimen in the background of the image. Then using fuzzy C-means clustering algorithm classifies the calculated Mahalanobis distances into two categories: the foreground and the background. Finally, using filling-hole technique improves the quality of the matting. This essay compares our algorithm with other Grow-Cut algorithms, Mahalanobis algorithm, KNN algorithm and Normalized Regression algorithm via experimental simulations. Experimental results show that the proposed algorithm can automatically segment the images with relative monotone color backgrounds in sound accuracy. Comparing with users' interaction methods, the novel algorithm is automatic, less time, and has lower complexity. For images with small random noises, segmentation results are also well looked. Also related to some specific problems, the algorithm get a better results too. In order to obtain quantitative evaluations to all the segmentation results, this essay put forward matting error to identify the results.
Keywords: Image matting, Mahalanobis distance, Fuzzy C-means clustering, Filling-holes.
Download PDF