Fazly Salleh Abas

An integrated content and metadata based retrieval system for art

A new approach to image retrieval is presented in the domain of museum and gallery image collections. Specialist algorithms, developed to address specific retrieval tasks, are combined with more conventional content and metadata retrieval approaches, and implemented within a distributed architecture to provide cross-collection searching and navigation in a seamless way. External systems can access the different collections using interoperability protocols and open standards, which were extended to accommodate content based as well as text based retrieval paradigms. After a brief overview of the complete system, we describe the novel design and evaluation of some of the specialist image analysis algorithms, including a method for image retrieval based on sub-image queries, retrievals based on very low quality images and retrieval using canvas crack patterns. We show how effective retrieval results can be achieved by real end-users consisting of major museums and galleries, accessing the distributed, but integrated, digital collections.

Craquelure analysis for content-based retrieval

In this paper, we describe a method for the extraction of distinguishable features from crack patterns, particularly those in paintings. First, we filter the selected crack image using 8 differently oriented Gabor filters. Then we thin the image to 1 pixel wide using a morphological thinning algorithm. Next we implement a crack following algorithm and generate statistical structure of global and local features from a chain code based representation. We describe an orientation-based feature extraction method to represent a crack network from sets of local orientation features. The resultant features are used as a guide towards classifying crack network patterns into several predefined classes, i.e circular, rectangular, spider-web, unidirectional and random. A simple classification experiment is presented to describe the significance of those extracted features towards classifying craquelure patterns.

Classification of painting cracks for content-based analysis

In this paper we present steps taken to implement a content-based analysis of crack patterns in paintings. Cracks are first detected using a morphological top-hat operator and grid-based automatic thresholding. From a 1-pixel wide representation of crack patterns, we generate a statistical structure of global and local features from a chain-code based representation. A well structured model of the crack patterns allows post-processing to be performed such as pruning and high-level feature extraction. High-level features are extracted from the structured model utilising information mainly based on orientation and length of line segments. Our strategy for classifying the crack patterns makes use of an unsupervised approach which incorporates fuzzy clustering of the patterns. We present results using the fuzzy k-means technique.

Supervised Locally Linear Embedding in face recognition

Locally Linear Embedding (LLE), which has recently emerged as a powerful face feature descriptor, suffers from a limitation. That is class-specific information of data is lacked of during face analysis. Thus, we propose a supervised LLE technique, known as class-label Locally Linear Embedding (cLLE), to overcome the problem. cLLE is able to discover the nonlinearity of high-dimensional face data by minimizing the global reconstruction error of the set of all local neighbors in the data set. cLLE utilizes user class-specific information in neighborhoods selection and thus preserves the local neighborhoods. Since the locality preservation is correlated to the class discrimination, the proposed cLLE is expected superior to LLE in face recognition. Experimental results on three face databases: ORL, AR and Yale databases, demonstrate that the proposed technique obtains better recognition performance than PCA and LLE.

TRANSLATION AND SCALE INVARIANTS OF HAHN MOMENTS

Hahn moments are a superset of Tchebichef and Krawtchouk moments. The formulation for Hahn moments is however comparably more complex than other moments. So far only research work on translation and scale invariants for Tchebichef moments has been presented but not on Hahn moments. In this paper, a moment normalization method to achieve translation and scale invariants of Hahn moments is proposed. This method applies the concept of mapping functions used in image normalization. The mapping functions, once determined, are plugged into the moment generating functions to generate moment invariants. The proposed method is simpler and flexible. Experimental results show that faster execution and more precise moment invariants can be achieved using the invariant generating functions.

Collaborative Support for Medical Data Mining in Telemedicine

An advance in todays information communication technology (ICT) has opened an opportunity for healthcare industry to enhance and improve the quality of their services. Decision makers are now able to make their decisions accurately and precisely with the helps of computer-based decision support system. Sharing of patients medical records is not a new approach. From paper based format to electronic format, patients medical records are shared among physicians, medical staffs and etc. Shared data might be carried manually from one department to another department or sent physically from hospital to hospital. With Internet technology, some medical data were sent via e-mail. These methods encountered many problems to patients and practitioners. Security and privacy issues, data lost, and delivery durations were some of the concerns that should be overcome. This paper will discussed an idea on how to overcome above mentioned issues and proposed a solution that can be served as the platform for future medical data sharing in telemedicine. The successful development of the working prototype will greatly enhance the functionality of existing data sharing in the hospital. At the same time, the tools and algorithms designed in this idea will helps to solve some of the data mining challenges

Automatic change detection of Belum-Temengor forested area using multitemporal SAR images

Two C-band RADARSAT synthetic aperture radar (SAR) images are used to automatically detect changes that occured in between year 2004 and 2008 due to extensive deforestation, at Gadong River area of Belum-Temengor, Northern region of West Malaysia. Changes are detected by automatically finding the best threshold value of the standard log-ratio image that is derived from the two multitemporal SAR images. Minimum error thresholding is performed using relative entropy method. The result is validated using the difference of two Normalised Difference Vegetation Index (NDVI) images generated from two SPOT-5 images.

Nuclear IHC enumeration: A digital phantom to evaluate the performance of automated algorithms in digital pathology

Automatic and accurate detection of positive and negative nuclei from images of immunostained tissue biopsies is critical to the success of digital pathology. The evaluation of most nuclei detection algorithms relies on manually generated ground truth prepared by pathologists, which is unfortunately time-consuming and suffers from inter-pathologist variability. In this work, we developed a digital immunohistochemistry (IHC) phantom that can be used for evaluating computer algorithms for enumeration of IHC positive cells. Our phantom development consists of two main steps, 1) extraction of the individual as well as nuclei clumps of both positive and negative nuclei from real WSI images, and 2) systematic placement of the extracted nuclei clumps on an image canvas. The resulting images are visually similar to the original tissue images. We created a set of 42 images with different concentrations of positive and negative nuclei. These images were evaluated by four board certified pathologists in the task of estimating the ratio of positive to total number of nuclei. The resulting concordance correlation coefficients (CCC) between the pathologist and the true ratio range from 0.86 to 0.95 (point estimates). The same ratio was also computed by an automated computer algorithm, which yielded a CCC value of 0.99. Reading the phantom data with known ground truth, the human readers show substantial variability and lower average performance than the computer algorithm in terms of CCC. This shows the limitation of using a human reader panel to establish a reference standard for the evaluation of computer algorithms, thereby highlighting the usefulness of the phantom developed in this work. Using our phantom images, we further developed a function that can approximate the true ratio from the area of the positive and negative nuclei, hence avoiding the need to detect individual nuclei. The predicted ratios of 10 held-out images using the function (trained on 32 images) are within ±2.68% of the true ratio. Moreover, we also report the evaluation of a computerized image analysis method on the synthetic tissue dataset.

Vision-Based Human Gesture Recognition Using Kinect Sensor

Gestures are indeed important in our daily life as they serve as one of the communication platform by using body motions in order to deliver information or effectively interact. This paper proposes to leverage the Kinect sensor for close-range human gesture recognition. The orientation details of human arms are extracted from the skeleton map sequences in order to form a bag of quaternions feature vectors. After the conversion to log-covariance matrix, the system is trained and the gestures are classified by multi-class SVM classifier. An experimental dataset of skeleton map sequences for 5 subjects with 6 gestures was collected and tested. The proposed system obtained remarkably accurate result with nearly 99 % of average correct classification rate (ACCR) compared to state of the art method with ACCR of 95 %.

Multi-level shape description technique

Image database querying has become a major area of research and received increased attention. Shape representation is one of the major problems in content-based image retrieval (CBIR). Shape representation generally looks for effective ways to capture the essence of the shape features that make it easier for shapes to be stored, recognized, compared against, and retrieved. In this paper, we shall present a structural shape representation and description technique, where object shape is represented at multiple levels of abstraction. This technique has been implemented in order to analyze its retrieval accuracy, the performance evaluation showed very remarkable and promising results.