Wei-Bang Chen

A PCA-Based Vehicle Classification Framework

Due to its great practical importance, Intelligent Transportation System has been an active research area in recent years. In this paper, we present a framework that incorporates various aspects of an intelligent transportation system with its ultimate goal being vehicle classification. Given a traffic video sequence, the proposed system first proceeds to segment individual vehicles. Then the extracted vehicle objects are normalized so that all vehicles are aligned along the same direction and measured at the same scale. Following the preprocessing step, two classification algorithms - Eigenvehicle and PCA-SVM, are proposed and implemented to classify vehicle objects into trucks, passenger cars, vans, and pick-ups. These two methods exploit the distinguishing power of Principal Component Analysis (PCA) at different granularities with different learning mechanisms. Experiments are conducted to compare these two methods and the results demonstrate the effectiveness of the proposed framework.

An automated bacterial colony counting and classification system

Bacterial colony enumeration is an essential tool for many widely used biomedical assays. However, bacterial colony enumerating is a low throughput, time consuming and labor intensive process since there may exist hundreds or thousands of colonies on a Petri dish, and the counting process is usually manually performed by well-trained technicians. In this paper, we introduce a fully automatic yet cost-effective bacterial colony counter which can not only count but also classify colonies. Our proposed method can recognize chromatic and achromatic images and thus can deal with both color and clear medium. In addition, the proposed method is software-centered and can accept general digital camera images as its input. The counting process includes detecting dish/plate regions, identifying colonies, separating aggregated colonies, and reporting colony counts. In order to differentiate colonies of different species, the proposed counter adopts one-class Support Vector Machine (SVM) with Radial Basis Function (RBF) as the classifier. Our proposed counter demonstrates a promising performance in terms of both precision and recall, and is robust and efficient in terms of labor-and time-savings.

A three-way investigation of a game-CAPTCHA: automated attacks, relay attacks and usability

Existing captcha solutions on the Internet are a major source of user frustration. Game captchas are an interesting and, to date, little-studied approach claiming to make captcha solving a fun activity for the users. One broad form of such captchas -- called Dynamic Cognitive Game (DCG) captchas -- challenge the user to perform a game-like cognitive task interacting with a series of dynamic images. We pursue a comprehensive analysis of a representative category of DCG captchas. We formalize, design and implement such captchas, and dissect them across: (1) fully automated attacks, (2) human-solver relay attacks, and (3) usability. Our results suggest that the studied DCG captchas exhibit high usability and, unlike other known captchas, offer some resistance to relay attacks, but they are also vulnerable to our novel dictionary-based automated attack.

An Automated Bacterial Colony Counting System

Bacterial colony enumeration is an essential tool for many widely used biomedical assays. However, bacterial colony enumerating is a low throughput, time consuming and labor intensive process since there might exist hundreds or thousands of colonies on a Petri dish, and the counting process is often manually performed by well-trained technicians. In this paper, we introduce a fully automatic yet cost-effective bacterial colony counter. Our proposed method can recognize chromatic and achromatic images and thus can deal with both color and clear medium. In addition, our proposed method can also accept general digital camera images as its input. The whole process includes detecting dish/plate regions, identifying colonies, separating aggregated colonies, and finally reporting consistent and accurate counting results. Our proposed counter has a promising performance in terms of both precision and recall, and is flexible and efficient in terms of labor- and time- savings.

An Automated Gridding and Segmentation Method for cDNA Microarray Image Analysis

Gridding and spot segmentation are two critical steps in microarray gene expression data analysis. However, the problems of noise contamination and donut-shaped spots often make signal extraction process a labor intensive task. In this paper, we propose a three-step method for automatic gridding and spot segmentation. The method starts with a background removal and noise eliminating step, and then proceeds in two steps. The first step applies a fully unsupervised method to extract blocks and grids from the cleaned data. The second step applies a simple, progressive spot segmentation method to deal with inner holes and noise in spots. We tested its performance on real microarray images against a widely used software GenePix. Our results show that the proposed method deals effectively with poor-conditioned microarray images in both gridding and spot segmentation

A Multiple Instance Learning and Relevance Feedback Framework for Retrieving Abnormal Incidents in Surveillance Videos

This paper incorporates coupled hidden Markov models (CHMM) with relevance feedback (RF) and multiple-instance learning (MIL) for retrieving various abnormal events in surveillance videos. CHMM is suitable for modeling not only the object’s behavior itself but also the interactions between objects. In addition, to address the challenges posed by the “semantic gap” between high level human concepts and the machine-readable low level visual features, we introduce relevance feedback (RF) to bridge the semantic gap by progressively collecting feedback from the user, which allows the machine to discover the semantic meanings of an event by exploring the patterns behind lowlevel features. The adopted multiple-instance learning algorithm enables the proposed framework to provide a user-friendly video retrieval platform with the use of queryby- example (QBE) interface. The experimental results show the effectiveness of the proposed framework in detecting “chasing”, “fighting”, and “robbery” events by demonstrating the increase of retrieval accuracy through iterations and comparing with other methods. By tightly integrating these key components in a learning system, we ease the surveillance video retrieval problem.

An improvement of color image segmentation through projective clustering

Image segmentation as the processing of partitioning a digital image into multiple segments has wide applications, such as image retrieval, medical inspection, and computer forensics. Clustering methods as one solution are applied on a single or multiple feature spaces of an image, such as color, intensity, or texture, in order to group similar pixels that share certain visual characteristics. Given a particular color image, not all features from a color space, such as RGB, HSV, or Lab, are equally effective in describing the visual characteristics of segments. In this paper, we propose a projective clustering algorithm HCPC (Hill-Climbing based Projective Clustering) which utilizes EPCH (an efficient projective clustering technique by histogram construction) as the main framework and hill-climbing algorithm for dense region detection, for color image segmentation, thereby finding interesting clusters (segments) within subspaces of a given feature space. A new feature space, named HSVrVgVb, is also explored which is derived from HSV (Hue, Saturation, and Value) color space. The experimental results show that compared with hill-climbing algorithm (for efficient color-based image segmentation), our proposed algorithm is more scalable when the dimensionality of feature space is high, and also generates comparable segmentation results.

Image spam clustering: an unsupervised approach

We propose an unsupervised image clustering framework for revealing the common origins, i.e. the spam gangs, of unsolicited emails. In particular, we target email spam with image attachments because spam information is harder to extract due to information hiding enabled by various image obfuscation techniques. To identify spam gangs, we observe that spam images from the same source are usually composed of visually similar elements which are arranged and altered in many different ways in order to trick the spam filter. We propose to infer spam images originated from the same spam gang by investigating spam email similarity in terms of their visual appearance and editing style. In particular, a data mining technique based on unsupervised image clustering is proposed in this paper to solve this problem. This is achieved by first dividing a spam image into different areas/segments, including texts, foreground graphic illustrations, and background areas. The proposed framework then extracts characteristic visual features from segmented areas, including text layout, visual features of foreground graphic illustrations and its spatial layout, and background texture features. In the clustering stage, all spam images are first categorized as illustrated images and text mainly images according to the existence of foreground illustration objects. Then illustrated images are clustered based on the color and/or foreground layout, while text mainly images are clustered based on the text layouts and/or background textures. A novel unsupervised ranked clustering algorithm is proposed for feature fusion, which is used in combination with the traditional hierarchical clustering algorithm for clustering. We test the proposed approach using different settings and combinations of features and measure the overall performance with V-measure.

Revealing common sources of image spam by unsupervised clustering with visual features

In this paper, we investigate image spam with data mining techniques in order to reveal the common sources of unsolicited emails. To identify the origins, a two-stage clustering method groups visually similar spam images by exploring their visual features, including color feature, layout feature, text layout, and background textures. We test the proposed approach under different settings and combinations of features and measure the performance with a modified F-measure.

Automatic in vivo microscopy video mining for leukocytes

Biological videos are very different from conventional videos. Automatic spatiotemporal mining of moving cells from in vivo microscopy videos is extremely difficult because of the severe noises, camera/subject movements, deformations, and strong dependencies on microscopy operators. In this paper, we present an automatic spatiotemporal mining system of rolling and adherent leukocytes for intravital videos. The magnitude of leukocyte adhesion and decrease in rolling velocity are common interests in inflammation response studies. Currently, there is no existing system which is perfect for such purposes. Several approaches have been proposed for tracking leukocytes. However, these approaches can either only track leukocytes that roll along the centerline of the blood vessel, or can only handle leukocytes with fixed morphologies. In addition, the camera/subject movement is a severe problem which occurs frequently while analyzing in vivo microscopy videos. In this paper, we proposed a new method for automatic recognition of non-adherent and adherent leukocytes. The proposed method includes three steps: (1) camera/subject movement alignment; (2) moving leukocytes detection; (3) adherent leukocytes detection. The experimental results demonstrate the effectiveness of the proposed method.