Shiming Ge

Learning multi-channel correlation filter bank for eye localization

Accurate eye localization plays a key role in many face analysis related applications. In this paper, we propose a novel statistic-based eye localization framework with a group of trained filter arrays called multi-channel correlation filter bank (MCCFB). Each filter array in the bank suits to a different face condition, thus combining these filter arrays can locate eyes more precisely in the conditions of variable poses, appearances and illuminations when comparing to single filter based or filter array based methods. To demonstrate the performance of our framework, we compare MCCFB with other statistic-based eye localization methods, experimental results show superiority of our method in detection ratio, localization accuracy and robustness.

Abnormal event detection via adaptive cascade dictionary learning

Detecting abnormal events plays an essential role in video content analysis and has received increasing attention in surveillance system. One of the major problems in abnormal event detection is the imbalanced classification issue due to the rare abnormal samples. Another problem is the difficulty of detecting anomalies within a reasonable amount of computation time. To address these problems, we propose an adaptive cascade dictionary learning framework for detecting the anomalies. The framework considers anomaly detection as an one-class classification problem with a cascade of dictionaries. Each stage of the cascade constructs an adaptive dictionary to detect the anomalies with costless least square optimization solution. The experiments on benchmark datasets demonstrate that the proposed method has a better performance while comparing with several state-of-the-art methods.

Color Image Encryption in CIE L*a*b* Space

To protect the contents of images in the mobile internet era during image storage and transmission, image encryption has achieved a tremendous success during the last decades. Currently, little attention has been paid to non-RGB color spaces such as HSV, YUV and L*a*b* color spaces in the color image encryption community. In this paper we use high level encryption schemes in more informative channels and low level encryption schemes in less informative channels. This paper is the first time to encrypt color image in CIE L*a*b* color space. First we convert RGB to L*a*b* color space. The 2D Arnold’s cat map followed by the 3D Lu chaotic map are conducted in the L* channel. The less complicated DNA coding and 1D logistic map based encryption scheme is leveraged in the a* and b* channels, which contain less information than that in the L* channel. The experimental results reveal that our method achieves similar results with the method that conducts the same scheme in each channel of the RGB color space, while consuming less time. In addition, our method can resistant several attacks such as brute-force attack, statistic attack, correlation attack.

An Image Encryption Algorithm Based on Zigzag Transformation and 3-Dimension Chaotic Logistic Map

An image encryption algorithm based on Zigzag transformation and 3-Dimension Logistic chaotic map by making use of the permutation-diffusion encryption structure is proposed. The algorithm consists of two parts: firstly, Zigzag transformation is used to scramble pixel position of the color image through three channels; then, 3-Dimension Logistic chaotic map is utilized to diffuse pixel values through three channels. To solve the problem of large computation and space overhead of color image encryption algorithm, the iterative chaotic sequences are used several times in the diffusion process to improve encryption efficiency. The key space of the algorithm is large enough to resist brute-force attack and simulation results show that it also has high key sensitivity, high encryption speed and the strong ability to resist exhaustive attack and statistical attack.

Eye localization based on correlation filter bank

Eye localization is a key step in many face analysis related applications. In this paper, we present a novel eye localization method based on a group of trained filters called correlation filter bank (CFB). We formulate the eye localization problem as an optimization problem with a well-defined cost function based on CFB. The CFB is trained with an EM-like adaptive clustering approach. The trained filter bank includes several discriminative filter templates, each of them suits to a different face condition from the others, thus can provide accurate eye localization ability for variable poses, appearances and illuminations. Simulation comparisons with cascade classifier-based method [1], traditional single correlation filter based methods [2][3] and pictorial structure model based method [4] demonstrates the superiority of the proposed method both in detection ratio and localization accuracy.

Detecting Masked Faces in the Wild with LLE-CNNs

Detecting faces with occlusions is a challenging task due to two main reasons: 1) the absence of large datasets of masked faces, and 2) the absence of facial cues from the masked regions. To address these two issues, this paper first introduces a dataset, denoted as MAFA, with 30, 811 Internet images and 35, 806 masked faces. Faces in the dataset have various orientations and occlusion degrees, while at least one part of each face is occluded by mask. Based on this dataset, we further propose LLE-CNNs for masked face detection, which consist of three major modules. The Proposal module first combines two pre-trained CNNs to extract candidate facial regions from the input image and represent them with high dimensional descriptors. After that, the Embedding module is incorporated to turn such descriptors into a similarity-based descriptor by using locally linear embedding (LLE) algorithm and the dictionaries trained on a large pool of synthesized normal faces, masked faces and non-faces. In this manner, many missing facial cues can be largely recovered and the influences of noisy cues introduced by diversified masks can be greatly alleviated. Finally, the Verification module is incorporated to identify candidate facial regions and refine their positions by jointly performing the classification and regression tasks within a unified CNN. Experimental results on the MAFA dataset show that the proposed approach remarkably outperforms 6 state-of-the-arts by at least 15.6%.

Traffic Sign Recognition Based on Attribute-Refinement Cascaded Convolutional Neural Networks

Traffic sign recognition is a critical module of intelligent transportation system. Observing that a subtle difference may cause misclassification when the actual class and the predictive class share the same attributes such as shape, color, function and so on, we propose a two-stage cascaded convolutional neural networks CNNs framework, called attribute-refinement cascaded CNNs, to train the traffic sign classifier by taking full advantage of attribute-supervisory signals. The first stage CNN is trained with class label as supervised signals, while the second stage CNN is trained on super classes separately according to auxiliary attributes of traffic signs for further refinement. Experiments show that the proposed hierarchical cascaded framework can extract the deep information of similar categories, improve discrimination of the model and increase classification accuracy of traffic signs.

Negative-Supervised Cascaded Deep Learning for Traffic Sign Classification

In this paper, we propose a novel deep learning framework for object classification called negative-supervised cascaded deep learning. There are two hierarchies in our cascaded method: the first one is a convolutional neural network trained on positive-only samples, which is used to select supervisory samples from a negative library. The second one is inherited from the trained first CNN. It is trained on positive and negative samples, which are selected from domain related database by utilizing negative-supervised mechanism. Experiments are applied this idea to traffic sign classification using two classic convolutional neural networks, LeNet-5 and AlexNet as baselines. Classification rates improved by \(0.7\%\) and \(1.1\%\) with LeNet-5 and AlexNet respectively, which demonstrates the efficiency and superiority of our proposed framework.

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.

Color image encryption in non-RGB color spaces

To protect the contents of images in the mobile internet era during image storage and transmission, image encryption has achieved a tremendous success during the last decades. Traditional color image encryption method often use the RGB color space. We have the observation that in non-RGB color spaces, the luminance channels often contain more information for content recognition than the chroma channels do. Thus, in this paper we propose to use high level encryption schemes in more informative channels and low level encryption schemes in less informative channels. The 2D Arnold’s cat map followed by the 3D Lu chaotic map are conducted in the luminance channel. The less complicated DNA coding and 1D logistic map based encryption scheme is leveraged in the chroma channels. We use this strategies in 4 typical non-RGB color spaces, i.e., YCbCr, YIQ, HSV, L*a*b*. We evaluate and compare the performances and the time consumptions of the methods in the 4 Non-RGB color spaces. The experimental results reveal that the encryption methods in Non-RGB color spaces can achieve similar results as the method that conducts the same encryption level in each channel of the RBG color space, including the resistance to several attacks such as brute-force attack, statistic attack, correlation attack, while consuming less time. The method in YCbCr color space performances the best in the time consumption.