Zhensong Chen

Automatic Road Crack Detection Using Random Structured Forests

Cracks are a growing threat to road conditions and have drawn much attention to the construction of intelligent transportation systems. However, as the key part of an intelligent transportation system, automatic road crack detection has been challenged because of the intense inhomogeneity along the cracks, the topology complexity of cracks, the inference of noises with similar texture to the cracks, and so on. In this paper, we propose CrackForest, a novel road crack detection framework based on random structured forests, to address these issues. Our contributions are shown as follows: 1) apply the integral channel features to redefine the tokens that constitute a crack and get better representation of the cracks with intensity inhomogeneity; 2) introduce random structured forests to generate a high-performance crack detector, which can identify arbitrarily complex cracks; and 3) propose a new crack descriptor to characterize cracks and discern them from noises effectively. In addition, our method is faster and easier to parallel. Experimental results prove the state-of-the-art detection precision of CrackForest compared with competing methods.

Image Segmentation via Improving Clustering Algorithms with Density and Distance

Abstract Image segmentation problem is a fundamental task and process in computer vision and image processing applications. It is well known that the performance of image segmentation is mainly influenced by two factors: the segmentation approaches and the feature presentation. As for image segmentation methods, clustering algorithm is one of the most popular approaches. However, most current clustering-based segmentation methods exist some problems, such as the number of regions of image have to be given prior, the different initial cluster centers will produce different segmentation results and so on. In this paper, we present a novel image segmentation approach based on DP clustering algorithm. Compared with the current methods, our method has several improved advantages as follows: 1) This algorithm could directly give the cluster number of the image based on the decision graph; 2) The cluster centers could be identified correctly; 3) We could simply achieve the hierarchical segmentation according to the applications requirement. A lot of experiments demonstrate the validity of this novel segmentation algorithm.

A novel clustering-based image segmentation via density peaks algorithm with mid-level feature

Image segmentation is an important and fundamental task in computer vision. Its performance is mainly influenced by feature representations and segmentation algorithms. In this paper, we propose a novel clustering-based image segmentation approach which can be called ICDP algorithm. It is able to capture the inherent structure of image and detect the nonspherical clusters. Compared to the other segmentation methods based on clustering, there are several advantages as follows: (1) Integral channel features are used to clearly and comprehensively represent the input image by naturally integrating heterogeneous sources of information; (2) cluster number can be determined directly and cluster centers are able to be identified automatically; (3) hierarchical segmentation is easy to be achieved via ICDP algorithm. The PSNR and MSE are applied to quantitatively evaluate the segmentation performance. Experimental results clearly demonstrate the effectiveness of our novel image segmentation algorithm.

Learning with label proportions based on nonparallel support vector machines

Learning a classifier from groups of unlabeled data, only knowing, for each group, the proportions of data with particular labels, is an important branch of classification tasks that are conceivable in many practical applications. In this paper, we proposed a novel solution for the problem of learning with label proportions (LLP) based on nonparallel support vector machines, termed as proportion-NPSVM, which can improve the classifiers to be a pair of nonparallel classification hyperplanes. The unique property of our method is that it only needs to solve a pair of smaller quadratic programming problems. Moreover, it can efficiently incorporate the known group label proportions with the latent unknown observation labels into one optimization model under a large-margin framework. Compared to the existing approaches, there are several advantages shown as follows: 1) it does not need to make restrictive assumptions on the training data; 2) nonparallel classifiers can be achieved without computing the large inverse matrices; 3) the optimization model can be effectively solved by using the alternative strategy with SMO technique or SOR method; 4) proportion-NPSVM has better generalization ability. Sufficient experimental results on both binary-classes and multi-classes data sets show the efficiency of our proposed method in classification accuracy, which prove the state-of-the-art method for LLP problems compared with competing algorithms.

Linear Twin SVM for Learning from Label Proportions

In this paper, we study the problem of learning from label proportions in which label information of data is provided in bag level. In this kind of problem, training data is grouped into various bags and only the proportions of positive instances is known. Inspired by proportion-SVM, we propose a new classification model based on twin SVM, which is also in a large-margin framework and only needs to solve two smaller problems. Avoiding making restrictive assumptions of the data, our model can learn the labels of every single instance based on group proportions information. In order to solve the non-convex problem in our new model, we propose an alternative algorithm to obtain the optimal solution efficiently. Also, we prove the effectiveness of our method in theoretical and experimental way.

Inverse extreme learning machine for learning with label proportions

In large-scale learning problem, the scalability of learning algorithms is usually the key factor affecting the algorithm practical performance, which is determined by both the time complexity of the learning algorithms and the amount of supervision information (i.e., labeled data). Learning with label proportions (LLP) is a new kind of machine learning problem which has drawn much attention in recent years. Different from the well-known supervised learning, LLP can estimate a classifier from groups of weakly labeled data, where only the positive/negative class proportions of each group are known. Due to its weak requirements for the input data, LLP presents a variety of real-world applications in almost all the fields involving anonymous data, like computer vision, fraud detection and spam filtering. However, even through the required labeled data is of a very small amount, LLP still suffers from the long execution time a lot due to the high time complexity of the learning algorithm itself. In this paper, we propose a very fast learning method based on inversing output scaling process and extreme learning machine, namely Inverse Extreme Learning Machine (IELM), to address the above issues. IELM can speed up the training process by order of magnitudes for large datasets, while achieving highly competitive classification accuracy with the existing methods at the same time. Extensive experiments demonstrate the significant speedup of the proposed method. We also demonstrate the feasibility of IELM with a case study in real-world setting: modeling image attributes based on ImageNet Object Attributes dataset.

Pavement Distress Detection Using Random Decision Forests

Pavement distress detection is a key technology to evaluate pavement surface and crack severity. However, there are many challenging problems when using pavement distress detection technology to do road maintenance, such as the inference of textured surroundings with similar intensity to the distresses, the existence of intensity inhomogeneity along the distresses and the requirement of real-time detection in practice. To address these problems, we propose a novel method for pavement distress detection based on random decision forests. By introducing the color gradient features at multiple scales commonly used in contour detection, we extend the feature set of traditional distress detection methods and get the represented crack with richer information. During the process of training, we apply a subsampling strategy at each node to maintain the diversity of trees. With this work, we finally solve all the three problems mentioned above. In addition, according to the characteristics of random decision forests, our method is easy to parallel and able to conduct real-time detection. Experimental results show that our approach is faster and more accurate than existing methods.

Multi-view Collective Tensor Decomposition for Cross-modal Hashing

Multimedia data available in various disciplines are usually heterogeneous, containing representations in multi-views, where the cross-modal search techniques become necessary and useful. It is a challenging problem due to the heterogeneity of data with multiple modalities, multi-views in each modality and the diverse data categories. In this paper, we propose a novel multi-view cross-modal hashing method named Multi-view Collective Tensor Decomposition (MCTD) to fuse these data effectively, which can exploit the complementary feature extracted from multi-modality multi-view while simultaneously discovering multiple separated subspaces by leveraging the data categories as supervision information. Our contributions are summarized as follows: 1) we exploit tensor modeling to get better representation of the complementary features and redefine a latent representation space; 2) a block-diagonal loss is proposed to explicitly pursue a more discriminative latent tensor space by exploring supervision information; 3) we propose a new feature projection method to characterize the data and to generate the latent representation for incoming new queries. An optimization algorithm is proposed to solve the objective function designed for MCTD, which works under an iterative updating procedure. Experimental results prove the state-of-the-art precision of MCTD compared with competing methods.

Laplacian SVM for Learning from Label Proportions

Proportion-SVM has been deeply studied due to its broad application prospects, such as modeling voting behaviors and spam filtering. However, the geometric information has been widely ignored. Thus, current methods usually show sensitivity to noises. To address these problems, in this paper, we combine the proportion learning framework with Laplacian term. We exploit the advantages of Laplacian term. Extensive experiments show the effectiveness of our method.

Supervised Object Boundary Detection Based on Structured Forests

Object boundary detection is an interesting and challenging topic in computer vision. Learning and combining the local, mid-level and high-level information play an important role in most of the recent approaches. However, few characteristics of a certain type of object are exploited. In this paper, we propose a novel supervised machine learning framework for object boundary detection, which makes use of the specific object features, such as boundary shape, directions and intensity. In the learning process, structured forest models are employed to tackle the high dimensional multi-class problem. Various experiment results show that our framework outperforms the competing models in the proposed data set, indicating that our framework is highly effective in modeling boundary for specific type of objects.