Kunhui Lin

Real-Time Eye Detection in Video Streams

A fast eye detection scheme for use in video streams rather than still images is presented in this paper. The temporal coherence of sequential frames was used to greatly improve the detection speed. First, the eye detector trained by AdaBoost algorithm is used to obtain the rough eye positions. Then these candidate positions are filtered by geometrical patterns of human eyes. The detected eye regions are then taken as the initial detecting window. After each frame is detected, the detecting window is updated. The experiments focused on video stream to exploit the benefits of our detector. In our experiments the mean detection rate was 92.73% for 320 times 240 resolution test videos, with a speed of 24.98 ms per frame. This speed is faster than previous research; however the detection rate does not dramatically decrease.

Research on Face Recognition Based on PCA

Principal components analysis (PCA) is a basic method widely used in face feature extraction and recognition. In order to overcome the shortcoming of absent consideration of the between-class information and the defect of the inconvenient update of the eigen-space in the traditional PCA method, this paper proposed a cluster-based feature projection method. The method enlarges the difference of samples in the different classes, while the difference of the same classes is reduced. Experimental results on ORL face database show that the method has higher correct recognition rate and higher recognition speeds than traditional PCA algorithm.

Estimation of human body shape and cloth field in front of a kinect

Abstract This paper describes an easy-to-use system to estimate the shape of a human body and his/her clothes. The system uses a Kinect to capture the human׳s RGB and depth information from different views. Using the depth data, a non-rigid deformation method is devised to compensate motions between different views, thus to align and complete the dressed shape. Given the reconstructed dressed shape, the skin regions are recognized by a skin classifier from the RGB images, and these skin regions are taken as a tight constraints for the body estimation. Subsequently, the body shape is estimated from the skin regions of the dressed shape by leveraging a statistical model of human body. After the body estimation, the body shape is non-rigidly deformed to fit the dressed shape, so as to extract the cloth field of the dressed shape. We demonstrate our system and the therein algorithms by several experiments. The results show the effectiveness of the proposed method.

A K-means clustering with optimized initial center based on Hadoop platform

With the explosive growth of data, the traditional clustering algorithms running on separate servers can not meet the demand. To solve the problem, more and more researchers implement the traditional clustering algorithms on the cloud computing platforms, especially for K-means clustering. But, few researchers pay attention to the K-means clustering structure, and most of researchers optimized the model of the cloud computing platform to raise the computing speed of K-means clustering. However the problem of instability caused by the random initial centers still exists. In this paper, we propose a K-means clustering algorithm with optimized initial centers based on data dimensional density. This method avoids the deficiency of the random initial centers and improves the stability of the K-means clustering. The experimental results show that the approach achieves a good performance on K-means, and improves the accuracy of K-means clustering on the test set.

A hybrid recommendation algorithm based on Hadoop

Recommender system has been widely used and collaborative filtering algorithm is the most widely used algorithm in recommender system. As scale of recommender system continues to expand, the number of users and items of recommender system is growing exponentially. As a result, the single-node machine implementing these algorithms is time-consuming and unable to meet the computing needs of large data sets. To improve the performance, we proposed a distributed collaborative filtering recommendation algorithm combining k-means and slope one on Hadoop. Apache Hadoop is an open-source organizations distributed computing framework. In this paper, the former hybrid recommendation algorithm was designed to parallel on MapReduce framework. The experiments were applied to the MovieLens dataset to exploit the benefits of our parallel algorithm. The experimental results present that our algorithm improves the performance.

Objective Classification Using Advanced Adaboost Algorithm

Adaboost, a general method for improving the accuracy of any given learning algorithm, is usually used to solve the problem of object detection based on cascade structure. However it has some disadvantage. The paper proposes an advanced Adaboost algorithm for object detection. The algorithm adopts a new method to update weighted parameters of weak classifiers. The weights are affected not only by the error rates, but also by their capacity of positive recognition. It is more adaptive to the object detection by decreasing the false alarm rates in the low false rejection rate terminal. The experiment results show the improvement achieved by the new algorithm.

An Intelligent TCM Diagnostic System Based on Intuitionistic Fuzzy Set

This paper proposes an intelligent traditional Chinese medical diagnostic system based on multi-agent system. In this system we also introduce an intelligent fuzzy diagnostic model based on intuitionistic fuzzy set theory according to the characteristics of traditional Chinese medicine, which realize intelligent diagnosis by modeling medical diagnosis rules via fuzzy relations, finally we propose a new approach for similarity measure between the intuitionistic fuzzy sets of syndromes.

Adaptive location recommendation algorithm based on location-based social networks

With the development of social network and location-based services, location-based social network rose. In the Geo-Social recommended system, location recommendation has become a focus of recent research. This paper analyzes three questions the personalized recommendation algorithm may face: location data sparseness, cold start and registered locations near and far from the usual residence. Through the analysis of those questions, we propose an improved adaptive location recommendation algorithm. This algorithm merges user collaborative filtering, social influence, and naive Bayesian classification. It adapts to the users current location, and recommend the most suitable location. In this paper, we compare the improved algorithm with other recommendation algorithms, verifying the feasibility, and effectiveness of the improved algorithm. Experimental results indicate that the improved algorithm can solve the problems of personalized place recommendations, and recommend place better.

A Fast Eye Localization Algorithm Using Integral Image

Eye localization is an essential step in automatic face recognition system(AFRC), since it has a direct influence on the overall recognition performance. In this paper, we present a simple and fast eye localization algorithm, called Integral Image Approach (IIA), for real time face recognition system. IIA can find out candidate positions for pupils and brows using the representation of image. Then by applying some constraints, such as the position of left pupil should be lower than that of the left brow, and adjusting the results, the IIA can eventually retrieve the locations of two pupils in a given face. The experiments show that the proposed IIA can locate eyes considerably fast with high accuracy in nearly frontal faces.

Domain-Specific Information Retrieval Based on Improved Language Model

There are two key ingredients in the general framework of language models used in information retrieval, one is importance weighting, the other is word relationship computing. A series of improvements are made to these ingredients of the general framework of language models which is used in domain-specific information retrieval. First, an EM algorithm is proposed to estimate the importance weights of query terms, and the Bayesian smoothing is used to compute the productive probabilities of important terms. Next, a new algorithm based on Dynamic Bayesian Network is proposed for obtaining the explanation probabilities between terms. Experiment shows that the improved model performs remarkably better for domain-specific information retrieval than some traditional retrieval techniques, and the extended framework has good expansibility.