Color Image Segmentation Using Proximal Classifier and Quaternion Radial Harmonic Fourier Moments.
Xiang-Yang Wang,Qian Wang,Xue-Bin Wang,Hong-Ying Yang,Zhi-Fang Wu,Pan-Pan Niu
DOI: https://doi.org/10.1007/s10044-019-00826-y
IF: 2.307
2019-01-01
Pattern Analysis and Applications
Abstract:Segmentation involves separating an object from the background in a given image. Image segmentation has a variety of applications and has received considerable attention in multimedia application and computer vision. Although numerous approaches have been introduced, image segmentation is still far from being solved due to most of image segmentation algorithms are often so complicated and some unsatisfactory results appear frequently. Therefore, developing a suitable technique of image segmentation is still a challenging problem. In this article, a novel color image segmentation will be introduced based on quaternion radial harmonic Fourier moments (QRHFMs) and proximal classifier. Firstly, the image feature of pixel-level is represented by the accurate and invariant QRHFMs holistically as a vector field, which can describe sufficiently the image pixel information due to take into account the relationship among different color channels. Secondly, the image feature from pixel-level is utilized as the input of the proximal classifier with consistency (PCC), which not only has lower computation time but also has better generalization compared to traditional support vector machines classifiers. Then, we choose the training samples by Tsallis entropy thresholding to train PCC classification model. Finally, the color image is classified by the trained PCC classification model. Our algorithm can make full use of the accurate and robust local image feature, as well the quickness and generalization ability of PCC classifier. A series of experimental results shows that this algorithm has better segmentation performance than the state-of-the-art method from the literature.