Huaxin Xiao

Data Separation of L1-minimization for Real-time Motion Detection

The `1-minimization used to seek the sparse solution restricts the applicability of compressed sensing. This paper proposes a data separation algorithm with computationally efficient strategies to achieve real-time performance of sparse model based motion detection. We use the traditional pursuit algorithms as a pre-process step that converts the iterative optimization into linear addition and multiplication operations. A novel motion detection method is implemented to compare the difference between the current frame and the background model in terms of sparse coefficients. The influence of dynamic texture or statistical noise diminishes after the process of sparse projection; thus, enhancing the robustness of the implementation. Results of the qualitative and quantitative evaluations demonstrate the higher efficiency and effectiveness of the proposed approach compared with those of other competing methods.

A Kalman Filter based Video Denoising Method Using Intensity and Structure Tensor

We propose a video denoising method based on Kalman filter to reduce the noise in video sequences. Firstly, with the strong spatiotemporal correlations of neighboring frames, motion estimation is performed on video frames consisting of previous denoised frames and current noisy frame based on intensity and structure tensor. The current noisy frame is processed in temporal domain by using motion estimation result as the parameter in the Kalman filter, while it is also processed in spatial domain using the Wiener filter. Finally, by weighting the denoised frames from the Kalman and the Wiener filtering, a satisfactory result can be obtained. Experimental results show that the performance of our proposed method is competitive when compared with state-of-the-art video denoising algorithms based on both peak signal-to-noise-ratio and structural similarity evaluations.

Salient object detection via robust dictionary representation

The theory of sparse and low-rank representation has worked competitive performance in the field of salient object detection. Generally, the salient object is represented as sparse error while the non-salient region is constrained by the property of low-rank. However, sparsity ignores the global structure which may break up the low-rank property. Besides, the outliers always lead to a poor representation. To handle these problems, this paper proposes a robust representation based on a discriminative dictionary which consists of non-salient and salient templates. Three weight measures are introduced and combined to select the proper templates. The coefficients on dictionary are restricted by ℓ2,1-norm. Correspondingly, Frobenius norm instead of ℓ1-norm is exploited to constrain the distribution of representation error. We compare the proposed algorithm against 17 state-of-the-art methods on 4 popular datasets by 6 evaluation metrics and demonstrate the competitive performance in terms of qualitative and quantitative results.

A Simplified Low Rank and Sparse Model for Visual Tracking

Object tracking is the process of determining the states of a target in consecutive video frames based on properties of motion and appearance consistency. Numerous tracking methods using low-rank and sparse constraints perform well in visual tracking. However, these methods cannot reasonably balance the two characteristics. Sparsity always pursues a sparse enough solution that ignores the low-rank structure and vice versa. Therefore, this paper replaces the low-rank and sparse constraints with 2,1 l norm. A simplified lowrank and sparse model for visual tracking (LRSVT), which is built upon the particle filter framework, is proposed in this paper. The proposed method first prunes particles which are different with the object and selects candidate particles for efficiency. A dictionary is then constructed to represent the candidate particles. The proposed LRSVT algorithm is evaluated against three related tracking methods on a set of seven challenging image sequences. Experimental results show that the LRSVT algorithm favorably performs against state-of-the-art tracking methods with regard to accuracy and execution time.

LASSO approximation and application to image super-resolution with CUDA acceleration

Sparse learning based methods are effective for image restoration applications since they make use of texture priors learned by pre-trained over-complete dictionaries. However, sparse learning based methods are extremely slow due to complexity of sparse decomposition and a large number of image patches to process. In this paper, we introduce a fast approximation for LASSO (Least Absolute Shrinkage and Selection Operator) and apply the approach to single image super-resolution with CUDA acceleration. Our approach utilizes linear combinations of pre-computed sparse codes of standard orthogonal bases to estimate the sparse code of input signal. Error analysis is performed to find the feasible conditions of our approach and upper-bound of the estimation error. The simplicity of our approach makes it easy to be implemented on GPU. As for super-resolution application, we apply our approach to improve one of the best super-resolution methods by Yang et. al. The super-resolution results are comparable with that of the state-of-the-art methods while the speed can be increased to 620%.

Data Based Color Constancy

Color constancy is an important task in computer vision. By analyzing the image formation model, color gamut data under one light source can be mapped to a hyperplane whose normal vector is only determined by its light source. Thus, the canonical light source is represented through the kernel method, which trains the color data. When an image is captured under an unknown illuminant, the image-corrected matrix is obtained through optimization. After being mapped to the high-dimensional space, the corrected color data are best fit for the hyperplane of the canonical illuminant. The proposed unsupervised feature-mining kernel method only depends on the color data without any other information. The experiments on the standard test datasets show that the proposed method achieves comparable performance with other state-of-the-art methods.

A robust motion detection algorithm on noisy videos

The applicability and performance of motion detection methods dramatically degrade with the increasing noise. In this paper, we propose a robust dictionary-based background subtraction approach, which formulates background modeling as a linear and sparse combination of atoms in a pre-learned dictionary. Motion detection is then implemented to compare the difference between sparse representations of the current frame and the background model. The projection of noise over the dictionary being irregular and random guarantees the adaptability of our approach. Experimental results on synthetic and real noisy videos demonstrate the robustness of the proposed approach compared to other methods.

A Low Illumination Environment Motion Detection Method based on Dictionary Learning

This paper proposes a dictionary-based motion detection method on video images captured under low light with serious noise. The proposed approach trains a dictionary by background images without foreground. It then reconstructs the test image according to the theory of sparse coding, and introduces the Structural Similarity Index Measurement (SSIM) as the detection standard to identify the detection caused by the brightness and contrast ratio changes. Experimental results show that compared to the mixture of Gaussian model and frame difference method, the proposed method can reach a better result under extreme low illumination circumstance.