Multi-Class Segmentation with Relative Location Prior

Multi-class image segmentation has made significant advances in recent years through the combination of local and global features. One important type of global feature is that of inter-class spatial relationships. For example, identifying “tree” pixels indicates that pixels above and to the sides are more likely to be “sky” whereas pixels below are more likely to be “grass.” Incorporating such global information across the entire image and between all classes is a computational challenge as it is image-dependent, and hence, cannot be precomputed. In this work we propose a method for capturing global information from inter-class spatial relationships and encoding it as a local feature. We employ a two-stage classification process to label all image pixels. First, we generate predictions which are used to compute a local relative location feature from learned relative location maps. In the second stage, we combine this with appearance-based features to provide a final segmentation. We compare our results to recent published results on several multi-class image segmentation databases and show that the incorporation of relative location information allows us to significantly outperform the current state-of-the-art.     

Nonparametric Bayesian Image Segmentation

Image segmentation algorithms partition the set of pixels of an image into a specific number of different, spatially homogeneous groups. We propose a nonparametric Bayesian model for histogram clustering which automatically determines the number of segments when spatial smoothness constraints on the class assignments are enforced by a Markov Random Field. A Dirichlet process prior controls the level of resolution which corresponds to the number of clusters in data with a unique cluster structure. The resulting posterior is efficiently sampled by a variant of a conjugate-case sampling algorithm for Dirichlet process mixture models. Experimental results are provided for real-world gray value images, synthetic aperture radar images and magnetic resonance imaging data.     

Partial Partitions,Partial Connections and Connective Segmentation

In connective segmentation (Serra in J. Math. Imaging Vis. 24(1):83–130, [2006]), each image determines subsets of the space on which it is “homogeneous”, in such a way that this family of subsets always constitutes a connection (connectivity class); then the segmentation of the image is the partition of space into its connected components according to that connection. Several concrete examples of connective segmentations or of connections on sets, indicate that the space covering requirement of the partition should be relaxed. Furthermore, morphological operations on partitions require the consideration of wider framework. We study thus partial partitions (families of mutually disjoint non-void subsets of the space) and partial connections (where connected components of a set are mutually disjoint but do not necessarily cover the set). We describe some methods for generating partial connections. We investigate the links between the two lattices of partial connections and of partial partitions. We generalize Serra’s characterization of connective segmentation and discuss its relevance. Finally we give some ideas on how the theory of partial connections could lead to improved segmentation algorithms.

A new graph cut-based multiple active contour algorithm without initial contours and seed points

This paper presents a new graph cut-based multiple active contour algorithm to detect optimal boundaries and regions in images without initial contours and seed points. The task of multiple active contours is framed as a partitioning problem by assuming that image data are generated from a finite mixture model with unknown number of components. Then, the partitioning problem is solved within a divisive graph cut framework where multi-way minimum cuts for multiple contours are efficiently computed in a top-down way through a swap move of binary labels. A split move is integrated into the swap move within that framework to estimate the model parameters associated with regions without the use of initial contours and seed points. The number of regions is also estimated as a part of the algorithm. Experimental results of boundary and region detection of natural images are presented and analyzed with precision and recall measures to demonstrate the effectiveness of the proposed algorithm.     

多模态特征融合的新闻视频故事分割方法

新闻视频故事分割是新闻视频分析的重要底层支持技术,本文提出了一种融合音频、视频等多模态特征的新闻视频故事分割方法.首先分析音频特征的静音片段作为音频特征候选点,对视频进行镜头分割,并将镜头分割结果分类为播音员镜头和新闻报道镜头,将所有的镜头分割点和播音员镜头片段提取为视频片段候选点;然后通过对新闻视频编辑规则的研究,对视频、音频特征候选点融合分析来获取新闻视频的故事分割,实验表明该方法在不同新闻视频编辑规则下都具有较好的分割效率.     

基于改进变分水平集的红外图像分割方法

提出一种基于水平集的红外图像偏微分分割方法,通过改进Chan-Vese模型中的能量函数获得偏微分方程,该能量函数将红外图像边缘与区域信息相结合,取得了全局极小值,该能量模型对水平集初始曲线的位置不敏感,并可定位图像边缘。基于该模型的变分水平集分割方法可分割出红外图像目标。实验结果表明,该方法效果良好,便于下一步的红外目标识别与跟踪。     

遥感图像分形特征提取与分割

分形理论由B.B.Mandelbrot于20世纪70年代中期创立,现已被广泛地应用于自然科学和社会科学的几乎所有领域.本文在前人研究的基础上,利用双毯法(Double Blanket Method)提取出图像的分形特征并用于图像分割,进一步证实了分形在此领域的可行性和有效性.首先,通过比较局部分形维数偏移全局分形维数的标准差来确定适合该方法的最优滑动窗口.其次,考虑到单尺度分维特征的局限性,提取出多尺度的特征值并建立分形维数谱.然后,以模拟图像为例,分析图像中各区域的分维谱,选择适当尺度的分形特征,利用最大似然法对图像进行分割.最后,将分形理论应用于遥感图像中,与传统的基于灰度值特征的图像分割方法相比,加入图像的空间分形纹理特征后分割精度明显提高.研究结果表明:分维值的大小和变化趋势可以表示不同地物的空间复杂度,结合地物的光谱以及灰度信息能有效地识别目标地物.     

路由器配置策略智能提取技术

为实现配置故障智能管理目标,提高路由器配置管理效率,提出对路由器配置文本按功能分段的算法,研究提取各分段配置策略的推理技术,构建基于可信度方法提取各分段配置策略的推理机模型。试验结果表明,它们均有94%以上的分段正确率和策略提取准确率,为构建配置故障智能管理系统做好了技术准备。     

聚类算法在银行客户细分中的应用

针对聚类算法在金融领域广泛应用的实际情况,基于银行客户数据集,对DBSCAN, K-means和X-means 3种聚类算法在执行效率、可扩展性、异常点检测能力等方面进行对比分析,并提出将X-means算法应用于银行业客户细分。利用X-means算法建立了一套银行客户细分模型,为银行决策者提供科学的决策支持。     

基于最小生成树聚类的中文版面分割法

针对中文版面多横竖混排的特点,提出一种基于最小生成树聚类的版面分割方法。对原图像进行水平和垂直游程平滑,并对平滑后所得的连通域进行预分类处理,将文本进行横排、竖排分类。对预分类后的各类文本采用最小生成树聚类算法进行聚类处理。经实验,准确率达97%。实验表明,该方法对中文文档有良好的分割效果。