Multi-Task Affinity Propagation Based Natural Image Matting

Image segmentation is an important research subject in the area of image processing. Most of the existing image segmentation methods partition the image based on the single cue of the image, the color, which brings a serious limitation when the complex scenes involve in the natural images. In this paper, we introduce a novel unsupervised image segmentation method via affinity propagation which takes into local texture and color features with superpixel map. The new method fuses color and texture information as local feature of each superpixel. The experimental results show that the proposed method performs better and steadier when partitioning various complex nature images, comparing to the existing methods.

Analysis of GC × GC fingerprints from medicinal materials using a novel contour detection algorithm: A case of Curcuma wenyujin

This paper presents a novel optimization method of training samples with Affinity Propagation (AP) clustering algorithm for multi-class Support Vector Machine (SVM) classification problem. The method of optimizing training samples is based on region clustering with affinity propagation algorithm. Then the multi-class support vector machines are trained for natural image classification with AP optimized samples. The feature space constructed in this paper is a composition of combined histogram with color, texture and edge descriptor of images. Experimental results show that better classification accuracy can be obtained by using the proposed method.

Focusing on the problem of natural image categorization, a novel multi-instance learning (MIL) algorithm based on affinity propagation (AP) clustering and support vector machine (SVM) is proposed. This algorithm regards every image as a bag, the low-level visual feature of each segmented region as instance. In order to transform every bag into a single sample, firstly, a collection of is generated by AP clustering method to construct a word-space. Secondly, according to the distance between visual-word and instance, a nonlinear mapping is defined to embed each bag as a coordinate point in the word-space, thereby obtaining every bag's mapping feature. As a result, the MIL problem is converted into a standard supervised learning problem, and then standard SVM classifiers are trained for image categorization in the word-space. Experimental results on the COREL data sets show that the proposed method, named APSVM-MIL, is robust, and its performance is superior to other key existing MIL algorithms.

Graph-based segmentation methods have become a major trend in computer vision. Due to the advantages of assimilating different graphs, a multi-scale fusion graph have a better performance than a single graph with single-scale. However, it is not reliable to determine a principle of graph combination. In this paper, we propose an adaptive affinity graph with subspace pursuit (AASP-graph) for natural image segmentation. The input image is first over-segmented into superpixels at different scales. An improved affinity propagation clustering method is proposed to select global nodes of these superpixels adaptively. Then, a L0-graph at each scale is obtained by a sparse representation of global nodes based on subspace pursuit. The adjacency-graph is finally built upon all superpixels of each scale and updated by the L0-graph. Experimental results on the Berkeley segmentation database show the effectiveness of the proposed AASP-graph in comparison with state-of-the-art approaches.

Image segmentation and region matching is the mainly issues of Region-Based Image Retrieval. In this paper we proposed adaptive image segmentation based on affinity propagation clustering (AAPSEG) and weighted region matching (WRM), APPSEG method calculated preference in Affinity Propagation clustering (AP) adaptively, and applied Adaptive Affinity Propagation clustering to nature image segmentation. WRM computed the weight between regions by location, size etc. It increased the influence of main region and decreased the secondary region. The Modified Hausdorff distance which was used to calculate similarity between quire image and candidate images effectively solved the problem of region matching in RBIR. The results of experiments demonstrated that the new method obviously improved precision in RBIR.