I-Fan Shen

Real-Time Dehazing for Image and Video

Outdoor photography and computer vision tasks often suffer from bad weather conditions, observed objects lose visibility and contrast due to the presence of atmospheric haze, fog, and smoke. In this paper, we propose a new method for real-time image and video dehazing. Based on a newly presented haze-free image prior - dark channel prior and a common haze imaging model, for a single input image, we can estimate the global atmospheric light and extract the scene objects transmission. To prevent artifacts, we refine the transmission using a cross-bilateral filter, and finally the haze-free frame can be restored by inversing the haze imaging model. The whole process is highly parallelized, and can be easily implemented on modern GPUs to achieve real-time performance. Comparing with existing methods, our approach provides similar or better results with much less processing time. The proposed method can be further used for many applications such as outdoor surveillance, remote sensing, and intelligent vehicles. In addition, rough depth information of the scene can be obtained as a by-product.

Supervised learning on local tangent space

A novel supervised learning method is proposed in this paper. It is an extension of local tangent space alignment (LTSA) to supervised feature extraction. First LTSA has been improved to be suitable in a changing, dynamic environment, that is, now it can map new data to the embedded low-dimensional space. Next class membership information is introduced to construct local tangent space when data sets contain multiple classes. This method has been applied to a number of data sets for classification and performs well when combined with some simple classifiers.

Dimension reduction of microarray data based on local tangent space alignment

We introduce the new nonlinear dimension reduction method: LTSA, in dealing with the difficulty of analyzing high-dimensional, nonlinear microarray data. Firstly, we analyze the applicability of the method and we propose the reconstruction error of LTSA. The method is tested on Iris data set and acute leukemias microarray data. The results show good visualization performance. And LTSA outperforms PCA on determining the reduced dimension. There is only subtle change in the clustering correctness after dimension reduction by LTSA. It is evident that application of nonlinear dimension reduction techniques could have a promising perspective in microarray data analysis.

Segmentation of discrete vector fields

In this paper, we propose an approach for 2D discrete vector field segmentation based on the Green function and normalized cut. The method is inspired by discrete Hodge decomposition such that a discrete vector field can be broken down into three simpler components, namely, curl-free, divergence-free, and harmonic components. We show that the Green function method (GFM) can be used to approximate the curl-free and the divergence-free components to achieve our goal of the vector field segmentation. The final segmentation curves that represent the boundaries of the influence region of singularities are obtained from the optimal vector field segmentations. These curves are composed of piecewise smooth contours or streamlines. Our method is applicable to both linear and nonlinear discrete vector fields. Experiments show that the segmentations obtained using our approach essentially agree with human perceptual judgement

Fast image segmentation using region merging with a k-Nearest Neighbor graph

A fast region merging method is proposed for solving the image segmentation problem. Rather than focusing on the global features of the image, our attention is drawn to local relationship between neighbor pixels with the goal that all similar pixels should be segmented in the same region. In this paper, the image segmentation problem is treated as a region merging procedure. To solve the problem, an initial oversegmentation is performed on the image and a k-Nearest Neighbor (k-NN) Graph whose vertexes denote regions is built. A new region similarity measure function is also proposed and the region similarity is assigned to the edge as its weight, which can make use of pixel intensity, edge feature, texture and so forth in a unit form. In k-NN graph, each vertex chooses exactly k nearest neighbors to connect. With it, the computation complexity of merging process can be reduced to O(tauN log2 N); here, tau denotes the number of nearest neighbor updates required at each iteration while N denotes the number of the initial regions. Implementation of the proposed algorithm is introduced, and some experiment results are given to prove our methodpsilas robustness and efficiency.

Image denoising through locally linear embedding

This paper presents a novel scheme for image denoising. In spite of the sophistication of recent schemes, most algorithms show outstanding performance under their assumption, but totally fail in general cases and produce artifacts or destroy fine structures. Inspired by recent manifold learning methods, especially the locally linear embedding (LLE), our method utilizes the underlying fact that image patches in noisy and denoised images construct manifolds with similar local geometry in these two distinct spaces. According to LLE, we characterize local geometry by measuring how an image patch represented by a feature vector can be reconstructed by its nearest neighbors in feature space. Besides using the training image patches to construct the embedding, we also propose to overlap the target denoised image patches to satisfy local compatibility and smoothness constraints. The experimental results show that our method is flexible with noise type and achieves state-of-the-art performance particularly in terms of preserving the fine structures.

Image Tangent Space for Image Retrieval

Image tangent space is actually high-level semantic space learned from low-level feature space by modified local tangent space alignment which was originally proposed for nonlinear manifold learning. Under the assumption that a data point in image space can be linearly approximated by some nearest neighbors in its local neighborhood, we develop a lazy learning method to locally approximate the optimal mapping function between image space and image tangent space. That is, the semantics of a new query image in image space can be inferred by the local approximation in its corresponding image tangent space. While Euclidean distance induced by the ambient space is often used to represent the difference between images, clearly, their natural distance is possibly different from Euclidean distance. Here, we compare three distance metrics: Chebyshev, Manhattan and Euclidean distances, and find that Chebyshev distance outperforms the other two in measuring the semantic similarity during retrieval. Experimental results show that our approach is effective in improving the performance of image retrieval systems

Finding dominant sets in microarray data.

Clustering allows us to extract groups of genes that are tightly coexpressed from Microarray data. In this paper, a new method DSF_Clust is developed to find dominant sets (clusters). We have preformed DSF_Clust on several gene expression datasets and given the evaluation with some criteria. The results showed that this approach could cluster dominant sets of good quality compared to kmeans method. DSF_Clust deals with three issues that have bedeviled clustering, some dominant sets being statistically determined in a significance level, predefining cluster structure being not required, and the quality of a dominant set being ensured. We have also applied this approach to analyze published data of yeast cell cycle gene expression and found some biologically meaningful gene groups to be dug out. Furthermore, DSF_Clust is a potentially good tool to search for putative regulatory signals.

Similarity measure for vector field learning

Vector data containing direction and magnitude information other than position information is different from common point data only containing position information. Those general similarity measures for point data such as Euclidean distance are not suitable for vector data. Thus, a novel measure must be proposed to estimate the similarity between vectors. The similarity measure defined in this paper combines Euclidean distance with angle and magnitude differences. Based on this measure, we construct a vector field space on which a modified locally linear embedding (LLE) algorithm is used for vector field learning. Our experimental results show that the proposed similarity measure works better than traditional Euclidean distance.

Supervised learning for classification

Supervised local tangent space alignment is proposed for data classification in this paper. It is an extension of local tangent space alignment, for short, LTSA, from unsupervised to supervised learning. Supervised LTSA is a supervised dimension reduction method. It make use of the class membership of each data to be trained in the case of multiple classes, to improve the quality of classification. Furthermore we present how to determine the related parameters for classification and apply this method to a number of artificial and realistic data. Experimental results show that supervised LTSA is superior for classification to other popular methods of dimension reduction when combined with simple classifiers such as the k-nearest neighbor classifier.