Roziati Zainuddin

Automatic Multilevel Medical Image Annotation and Retrieval

Image retrieval at the semantic level mostly depends on image annotation or image classification. Image annotation performance largely depends on three issues: (1) automatic image feature extraction; (2) a semantic image concept modeling; (3) algorithm for semantic image annotation. To address first issue, multilevel features are extracted to construct the feature vector, which represents the contents of the image. To address second issue, domain-dependent concept hierarchy is constructed for interpretation of image semantic concepts. To address third issue, automatic multilevel code generation is proposed for image classification and multilevel image annotation. We make use of the existing image annotation to address second and third issues. Our experiments on a specific domain of X-ray images have given encouraging results.

Design of a Fuzzy-based Decision Support System for Coronary Heart Disease Diagnosis

In the present paper, a fuzzy rule-based system (FRBS) is designed to serve as a decision support system for Coronary heart disease (CHD) diagnosis that not only considers the decision accuracy of the rules but also their transparency at the same time. To achieve the two above mentioned objectives, we apply a multi-objective genetic algorithm to optimize both the accuracy and transparency of the FRBS. In addition and to help assess the certainty and the importance of each rule by the physician, an extended format of fuzzy rules that incorporates the degree of decision certainty and importance or support of each rule at the consequent part of the rules is introduced. Furthermore, a new way for employing Ensemble Classifiers Strategy (ECS) method is proposed to enhance the classification ability of the FRBS. The results show that the generated rules are humanly understandable while their accuracy compared favorably with other benchmark classification methods. In addition, the produced FRBS is able to identify the uncertainty cases so that the physician can give a special consideration to deal with them and this will result in a better management of efforts and tasks. Furthermore, employing ECS has specifically improved the ability of FRBS to detect patients with CHD which is desirable feature for any CHD diagnosis system.

Phonetically rich and balanced speech corpus for Arabic speaker-independent continuous automatic speech recognition systems

This paper describes an efficient framework for designing and developing Arabic speaker-independent continuous automatic speech recognition systems based on a phonetically rich and balanced speech corpus. The speech corpus contains 415 sentences recorded by 42 (21 male and 21 female) Arabic native speakers from 11 Arab countries representing three major regions (Levant, Gulf, and Africa). The developed system is based on the Carnegie Mellon University (CMU) Sphinx tools. The Cambridge HTK tools were also used in some testing stages. The speech engine uses 3-emitting state Hidden Markov Models (HMM) for tri-phone based acoustic models. Based on experimental analysis of 4.07 hours of training speech data, the acoustic model used continuous observations probability model of 16 Gaussian mixture distributions and the state distributions were tied to 400 senons. The language model contains both bi-grams and tri-grams. The system obtained 91.23% and 92.54% correct word recognition with and without diacritical marks respectively.

Parallel versus Perpendicular Plots: A Comparative Study

A comparison study between parallel and perpendicular (scatter plot) plot visualization techniques are addressed. Comparison discussion focuses on the disadvantage and advantages of the two techniques and a GOMS analysis of the related tasks based on a system called the PaCQ interface.

Suppressing False Nagatives in Skin Segmentation

Human skin segmentation in colored images is closely related to face detection and recognition systems as preliminary required step. False negative errors degrade segmentation accuracy and therefore considered as critical problem in image segmentation. A general innovative approach for human skin segmentation that substantially suppresses false negative errors has been developed. This approach employed multi-skin models using HSV color space. Four skin color clustering models were used, namely: standard-skin model, shadow-skin model, light-skin model, and redness-skin model. The color information was used to segment skin-like regions by transforming the 3-D color space to 2-D subspace. A rule-based classifier produces four skin-maps layers. Each layer reflects its skin model. Pixel-based segmentation and region-based segmentation approaches has been combined to enhance the accuracy. The inspiring results obtained show that the suppression of false negatives is substantial and leads to better detection and recognition.

Multi Skin Color Clustering Models for Face Detection

Automatic face detection in colored images is closely related to face recognition systems, as a preliminary critical required step, where it is necessary to search for the precise face location. We propose a reliable approach for skin color segmentation to detect human face in colored images under unconstrained scene conditions that overcoming the sensitivity to the variation in face size, pose, location, lighting conditions, and complex background. Our approach is based on building multi skin color clustering models using HSV color space, multi-level segmentation, and rule-based classifier. We proposed to use four skin color clustering models instead of single skin clustering model, namely: standard-skin model, shadow-skin model, light-skin model, high-red-skin model. We made an independent skin color clustering models by converting 3-D color space to 2-D without losing color information in order to find the classification boundaries for each skin color pattern class in 2-D. Once we find the classification boundaries, we process the input image with the first-level skin-color segmentation to produce four layers; each layer reflecting its skin-color clustering model. Then an iterative rule-based region grow is performed to create one solid region of interest which is presumed to be a face candidate region that will be passed to the second-level segmentation. In this approach we combine pixel-based segmentation and region-based segmentation using the four skin layers. We also propose skin-color correction (skin lighting) at shadow-skin layer to improve detection rate. In the second-level segmentation we use gray scale to segment the face candidate region into the most significant features using thresholding. Next step is to compute the X-Y-reliefs to locate the accurate position of facial features in each face candidate region and match it with our geometrical knowledge in order to classify the face candidate region to a face or non-face region. We present experimental results of our implementation and demonstrate the feasibility of our approach to be general purpose skin color segmentation for face detection problem.

Prosody generation by integrating rule and template-based approaches for emotional Malay speech synthesis

This paper presents a hybrid technique to enhance the quality of the rule-based approach to generate prosody for Malay speech synthesis by integrating prosody parametric manipulation with template parametric manipulation so as to increase the intonation variability of the synthesized output. Basically the prosodic features of the neutral synthesized speech are manipulated in an attempt to express the four basic emotions, namely happiness, anger, sadness and fear. We also present an objective methodology to evaluate the effectiveness of the synthesized output to generate the appropriate prosody in order to confirm the subjective perception tests.

Multi-Modality Medical Images Feature Analysis

In this research study we analyze visual features of texture, shape and color in multi-modality medical images. The analysis consists of contrast, correlation, energy and homogeneity in texture extraction. We apply Hu moment invariant method to extract shape features. Both texture and shape features are extract in local level. Color histogram is used in represent color descriptor in analysing global medical images. These features are then will be classified based on its modality using support vector machine classifier. It shows that different modality have different characteristic and the importance of selecting significance features.

A Transparent Fuzzy Rule-Based Clinical Decision Support System for Heart Disease Diagnosis

Heart disease (HD) is a serious disease and its diagnosis at early stage remains a challenging task. A well-designed clinical decision support system (CDSS), however, that provides accurate and understandable decisions would effectively help the physician in making an early and appropriate diagnosis. In this study, a CDSS for HD diagnosis is proposed based on a genetic-fuzzy approach that considers both the transparency and accuracy of the system. Multi-objective genetic algorithm is applied to search for a small number of transparent fuzzy rules with high classification accuracy. The final fuzzy rules are formatted to be structured, informative and readable decisions that can be easily checked and understood by the physician. Furthermore, an Ensemble Classifier Strategy (ECS) is presented in order to enhance the diagnosis ability of our CDSS by supporting its decision, in the uncertain cases, by other well-known classifiers. The results show that the proposed method is able to offer humanly understandable rules with performance comparable to other benchmark classification methods.

A Framework of Hybrid Semantic Speech Query via Stemmer for Quran Documents Results

Speech recognition has provided limited converted speech into text as query for cross language information retrieval (CLIR). Therefore, in this study, speech recognition to be integrated with cross language information retrieval system to investigate the Quran system results performance using semantic and stemmers as stemming semantic spoken query (SSSQ). The query however is based on the speech to be input and converted into text. Therefore, this study is conducted with the purposes to investigate the integration speech, semantic and stemmers approach against the queries and vice versa. Furthermore, it is also conducted to investigate the performance query based on total retrieve and relevant. The retrieval however, included the irrelevant documents because of the translation polysemy. Results from the experiments suggest that SSSQ is most important process in CLIR. It also found that semantic speech approach with stemmers contributes to better approach in using speech instead of keyboard to investigate its performance for Quran document results.