Zhili Zhou

Multiple dictionary pairs learning and sparse representation-based infrared image super-resolution with improved fuzzy clustering

Exploring sparse representation to enhance the resolution of infrared image has attracted much attention in the last decade. However, conventional sparse representation-based super-resolution aim at learning a universal and efficient dictionary pair for image representation. However, considering that a large number of different structures exist in an image, it is insufficient and unreasonable to present various image structures with only one universal dictionary pair. In this paper, we propose an improved fuzzy clustering and weighted scheme reconstruction framework to solve this problem. Firstly, the training patches are divided into multiple clusters by joint learning multiple dictionary pairs with improved fuzzy clustering method. The goal of joint learning is to learn the multiple dictionary pairs which could collectively represent all the training patches with smallest reconstruction error. So that the learned dictionary pairs are more precise and mutually complementary. Then, high-resolution (HR) patches are estimated according to several most accurate dictionary pairs. Finally, these estimated HR patches are integrated together to generate a final HR patch by a weighted scheme. Numerous experiments demonstrate that this framework outperforms some state-of-art super-resolution methods in both quantitatively and perceptually.

Multi-sensor image super-resolution with fuzzy cluster by using multi-scale and multi-view sparse coding for infrared image

Super-resolution (SR) methods are effective for generating a high-resolution image from a single low-resolution image. However, four problems are observed in existing SR methods. (1) They cannot reconstruct many details from a low-resolution infrared image because infrared images always lack detailed information. (2) They cannot extract the desired information from images because they do not consider that images naturally come at different scales in many cases. (3) They fail to reveal different physical structures of low-resolution patch because they extract features from a single view. (4) They fail to extract all the different patterns because they use only one dictionary to represent all patterns. To overcome these problems, we propose a novel SR method for infrared images. First, we combine the information of high-resolution visible light images and low-resolution infrared images to improve the resolution of infrared images. Second, we use multiscale patches instead of fixed-size patches to represent infrared images more accurately. Third, we use different feature vectors rather than a single feature to represent infrared images. Finally, we divide training patches into several clusters, and multiple dictionaries are learned for each cluster to provide each patch with a more accurate dictionary. In the proposed method, clustering information for low-resolution patches is learnt by using fuzzy clustering theory. Experiments validate that the proposed method yields better results in terms of quantization and visual perception than the state-of-the-art algorithms.

A new keypoint-based copy-move forgery detection for small smooth regions

Copy-move forgery is one of the most common types of image forgeries, where a region from one part of an image is copied and pasted onto another part, thereby concealing the image content in the latter region. Keypoint based copy-move forgery detection approaches extract image feature points and use local visual features, rather than image blocks, to identify duplicated regions. Keypoint based approaches exhibit remarkable performance with respect to computational cost, memory requirement, and robustness. But unfortunately, they usually do not work well if smooth background areas are used to hide small objects, as image keypoints cannot be extracted effectively from those areas. It is a challenging work to design a keypoint-based method for detecting forgeries involving small smooth regions. In this paper, we propose a new keypoint-based copy-move forgery detection for small smooth regions. Firstly, the original tampered image is segmented into nonoverlapping and irregular superpixels, and the superpixels are classified into smooth, texture and strong texture based on local information entropy. Secondly, the stable image keypoints are extracted from each superpixel, including smooth, texture and strong texture ones, by utilizing the superpixel content based adaptive feature points detector. Thirdly, the local visual features, namely exponent moments magnitudes, are constructed for each image keypoint, and the best bin first and reversed generalized 2 nearest-neighbor algorithm are utilized to find rapidly the matching image keypoints. Finally, the falsely matched image keypoints are removed by customizing the random sample consensus, and the duplicated regions are localized by using zero mean normalized cross-correlation measure. Extensive experimental results show that the newly proposed scheme can achieve much better detection results for copy-move forgery images under various challenging conditions, such as geometric transforms, JPEG compression, and additive white Gaussian noise, compared with the existing state-of-the-art copy-move forgery detection methods.

Retinex-based image enhancement framework by using region covariance filter

Clear images are critical in understanding real scenarios. However, the quality of images may be severely declined due to terrible conditions. Images exposed to such conditions are usually of low contrast, contain much noise, and suffer from weak details. And these drawbacks tend to negatively influence the subsequent processing tasks. Many existing image enhancement methods only solve a certain aspect of aforementioned drawbacks. This paper proposes a Retinex-based image enhancement framework that can increase contrast, eliminate noise, and enhance details at the same time. First, we utilize a region covariance filter to estimate the illumination accurately at multiple scales. The corresponding reflectance can be predicted by dividing the original image by its illumination. Second, we utilize contrast-limited adaptive histogram equalization to enhance the global contrast of original images because the illumination contains the low-frequency component. Third, since the reflectance contains the details of the original image and noise, we adopt a non-local means filter to eliminate noise and use a guided filter to enhance the details in the reflectance. Fourth, we synthesize the final enhanced image by fusing the enhanced illumination and reflectance at each scale. Experiments have proved the improvement of the proposed framework in terms of both visual perception and quantitative comparisons with other compared methods.

Multi-scale image fusion through rolling guidance filter

Abstract Image fusion is essential in enhancing visual quality by blending complementary images, which are derived from different captured conditions or different sensors in the same scene. The role of image fusion in the Internet of Things has become considerably important in the future. For instance, data captured by multiple visual sensors need further computation or fusion, which is based on a network of making a decision or an analysis. A new image fusion method is proposed by using rolling guidance filter and joint bilateral filter in this paper. First, the saliency maps of two source images are extracted by the Kirsch operator. Subsequently, the two source images are decomposed by rolling guidance filter to obtain multi-scale images. Second, joint bilateral filter and optimal correction are utilized to optimize the saliency maps and obtain the final weight maps. Finally, two fusion rules are used to restore the final fused image. The proposed method not only preserves the details of source images, but also suppresses the artifacts effectively. Experimental results prove that our method generates better effects on both visual perception and objective quantization than traditional methods.

MCRS: A course recommendation system for MOOCs

With the popularization development of MOOC platform, the number of online courses grows rapidly. Efficient and appropriate course recommendation can improve learning efficiency. Traditional recommendation system is applied to the closed educational environment in which the quantity of courses and users is relatively stable. Recommendation model and algorithm cannot directly be applied to MOOC platform efficiently. With the light of the characteristics of MOOC platform, MCRS proposed in this paper has made great improvement in the course recommendation model and recommendation algorithm. MCRS is based on distributed computation framework. The basic algorithm of MCRS is distributed association rules mining algorithm, which based on the improvement of Apriori algorithm. In addition, it is useful to mine the hidden courses rules in course enrollment data. Firstly, the data is pre-processed into a standard form by Hadoop. It aims to improve the efficiency of the basic algorithm. Then it mines association rules of the standard data by Spark. Consequently, course recommendation information is transferred into MySQL through Sqoop, which makes timely feedback and improves user’s courses retrieval efficiency. Finally, to validate the efficiency of MCRS, a series of experiments are carried out on Hadoop and Spark, and the results shows that MCRS is more efficient than traditional Apriori algorithm and Apriori algorithm based on Hadoop, and the MCRS is suitable for current MOOC platform.

Enhancing heterogeneous similarity estimation via neighborhood reversibility

With the popularity of social networks, people can easily generate rich content with multiple modalities. How to effectively and simply estimate the similarity of multi-modal content is becoming more and more important for providing better information searching service of rich media. This work attempts to enhance the similarity estimation so as to improve the accuracy of multi-modal data searching. Toward this end, a novel multi-modal feature extraction approach, which involves the neighborhood reversibility verifying of information objects with different modalities, is proposed to build reliable similarity estimation among multimedia documents. By verifying the neighborhood reversibility in both single- and multi-modal instances, the reliability of multi-modal subspace can be remarkably improved. In addition, a new adaptive strategy, which fully employs the distance distribution of returned searching instances, is proposed to handle the neighbor selection problem. To further address the out-of-sample problem, a new prediction scheme is proposed to predict the multi-modal features for new coming instances, which is essentially to construct an over-complete set of bases. Extensive experiments demonstrate that introducing the neighborhood reversibility verifying can significantly improve the searching accuracy of multi-modal documents.

Pairwise IPVO-based reversible data hiding

Recently, Peng et al. proposed a reversible data hiding method based on improved pixel-value-ordering (PVO) and prediction-error expansion. In this paper, a novel method is proposed by extending Peng et al.’s work. In our method, three largest (or smallest) pixels in a block are utilized to generate two differences, and a new pixel modification strategy is proposed so that the PVO remains unchanged after data embedding. Taking three largest pixels for example, we utilize the third largest pixel to predict the second largest one, and meanwhile use the second largest one to predict the maximum. In this way, two differences are obtained. They are modified jointly so as to be embedded with log 23 bits instead of 2 bits in the traditional RDH methods. The advantage of doing so is to exclude situations where PVO is changed. Moreover, two embedding layers are utilized together to further decrease the embedding distortion. Extensive experiments verify that the proposed method outperforms Peng et al. ’s and some other state-of-the-art works.

A novel scheme for infrared image enhancement by using weighted least squares filter and fuzzy plateau histogram equalization

High-quality thermal infrared (IR) images are always preferred in numerous real-world applications. However, acquired IR images, which have low contrast and signal-to-noise ratio (SNR) among other characteristics, have inferior quality because of various factors. To improve the quality of IR images, three main aspects must be addressed: global contrast, local contrast, and noise of IR images. Most of the existing methods focus only on some of these issues. In this paper, we propose a novel scheme to solve the three issues. First, an edge-preserving filter called weighted least squares filter is adopted to decompose an IR image into a low-frequency (LF) component and a sequence of high-frequency (HF) components. We propose a fuzzy plateau histogram equalization for the LF component to improve global contrast. A strategy is exploited to alter the coefficients of the HF components to enhance local contrast. The primitive result is synthesized with the enhanced LF and HF components. To suppress the noise in the primitive result, nonlocal means filter is applied to derive the final result. Numerous experiments are conducted. Experimental results demonstrate that the proposed scheme exhibits the best performance compared with the other methods.

Robust copy-move forgery detection based on multi-granularity Superpixels matching

In this paper, we propose a new multi-granularity superpixels matching based algorithm for the accurate detection and localization of copy-move forgeries, which integrated the advantages of keypoint-based and block-based forgery detection approaches. Firstly, we divide the original tempted image into non-overlapping and irregular coarse-granularity superpixels, and the stable image keypoints are extracted from each coarse-granularity superpixel. Secondly, the superpixel features, which is quaternion exponent moments magnitudes, are extracted from each coarse-granularity superpixel, and we find the matching coarse-granularity superpixels (suspected forgery region pairs) rapidly using the Exact Euclidean Locality Sensitive Hashing (E2LSH). Thirdly, the suspected forgery region pairs are further segmented into fine-granularity superpixels, and the matching keypoints within the suspected forgery region pairs are replaced with the fine-granularity superpixels. Finally, the neighboring fine-granularity superpixels are merged, and we obtain the detected forgery regions through morphological operation. Compared with the state-of-the-art approaches, extensive experimental results, conducted on the public databases available online, demonstrate the good performance of our proposed algorithm even under a variety of challenging conditions.