Dzulkifli Mohamad

Exploiting Voronoi diagram properties in face segmentation and feature extraction

Segmentation of human faces from still images is a research field of rapidly increasing interest. Although the field encounters several challenges, this paper seeks to present a novel face segmentation and facial feature extraction algorithm for gray intensity images (each containing a single face object). Face location and extraction must first be performed to obtain the approximate, if not exact, representation of a given face in an image. The proposed approach is based on the Voronoi diagram (VD), a well-known technique in computational geometry, which generates clusters of intensity values using information from the vertices of the external boundary of Delaunay triangulation (DT). In this way, it is possible to produce segmented image regions. A greedy search algorithm looks for a particular face candidate by focusing its action in elliptical-like regions. VD is presently employed in many fields, but researchers primarily focus on its use in skeletonization and for generating Euclidean distances; this work exploits the triangulations (i.e., Delaunay) generated by the VD for use in this field. A distance transformation is applied to segment face features. We used the BioID face database to test our algorithm. We obtained promising results: 95.14% of faces were correctly segmented; 90.2% of eyes were detected and a 98.03% detection rate was obtained for mouth and nose.

Vision-Based Sign Language Classification: A Directional Review

ABSTRACT Sign language is a communication tool for deaf and dumb people that includes known signs or body gestures to transfer meanings. It uses shapes, directions, movements of hands, and also facial expressions. A sign not only transmits a word but also conveys a tone. Many of the deaf people are not only able to speak, but also not able to write or read a language, so developing sign language translation or in other words sign language recognition (SLR) system can be very vital in their life. The SLR is extremely desired because of its capability to overcome the obstacles between deaf and normal people. It is one of the most important research fields in the human computer interaction studies. Hence, this paper presents an overview of the recent main research works with the vision-based SLR system, and the existing recognition techniques are discussed. Next, we focus on video-based SLR system and perform continuous SLR within video sequences.

Computer Generated Caricature: A Survey

Caricature is a pictorial representation of a person or subject in summarizing way by exaggerating the most distinctive features and simplifies the common features in order to make that subject different from others and at the same time, preserve the likeness of the subject. Computer Generated Caricature is developed in order to assist the user in producing caricature automatically or semi-automatically. It is derived from the rapid advance in computer graphics and computer vision and introduced as a part of computer graphics’ non-photo realistic rendering technologies as well. Recently, Computer Generated Caricature becomes particularly interesting research topic due to the advantageous features of privacy, security, simplification, amusement and their explosive emergent real-world application such as in magazine, digital entertainment, Internet and mobile application. On the basis of the previous facts, this paper surveys the uses of caricature in variety of applications, theories and rules in the art of drawing caricature, how these theories are simulated in the development of caricature generation system and the current research trend in this field. Computer generated caricature can be divided into two main categories based on their input data type: human centered approach and image processing approach. Next, process of generating caricature from input photo is explained briefly. It also reported the state of the art techniques in generating caricature by classifying it into four approaches: interactive, regularity-based, learning-based and predefined database of caricature illustration. Lastly, this paper will discuss relevant issues, problems and several promising direction of future research.

Neural network speaker dependent isolated Malay speech recognition system: handcrafted vs genetic algorithm

We compare two approaches in selecting neural network learning parameters and architecture. Traditionally they are found by trial and error (handcrafted) and alternatively, can be found using a genetic algorithm. Trial and error can find good solutions but the drawback is this method is time consuming and it can only try a few possible solutions while the genetic algorithm is known to be able to search for a good solution intelligently and faster with greater diversity of possible solutions. We tested the approaches on ten isolated Malay digits from 0 to 9. Three factors are compared between the two approaches: time to get a good solution; network learning convergence; and the recognition rate. Our findings show that the neural network using the genetic algorithm achieved 94% recognition rate while the handcrafted neural network achieved 95%. However, using the genetic algorithm, a good solution can be found within days while with the handcrafted method it took weeks. The network learning convergence for both approaches were relatively the same.

Content-based image retrieval using PSO and k-means clustering algorithm

In various application domains such as website, education, crime prevention, commerce, and biomedicine, the volume of digital data is increasing rapidly. The trouble appears when retrieving the data from the storage media because some of the existing methods compare the query image with all images in the database; as a result, the search space and computational complexity will increase, respectively. The content-based image retrieval (CBIR) methods aim to retrieve images accurately from large image databases similar to the query image based on the similarity between image features. In this study, a new hybrid method has been proposed for image clustering based on combining the particle swarm optimization (PSO) with k-means clustering algorithms. It is presented as a proposed CBIR method that uses the color and texture images as visual features to represent the images. The proposed method is based on four feature extractions for measuring the similarity, which are color histogram, color moment, co-occurrence matrices, and wavelet moment. The experimental results have indicated that the proposed system has a superior performance compared to the other system in terms of accuracy.

Deductive method for recognition of on-line handwritten Persian/Arabic characters

The choice of relevant techniques in preprocessing, segmentation and feature extraction is very efficient and effective in rate of online handwriting recognition system. This paper presents a novel deductive method for detecting critical points of the Persian/Arabic handwritten character system in all their different shapes. The implemented method has increased the performance rate of the online Persian/Arabic handwritten recognition system and has decreased the computational mistake for finding critical points. This method helps us to extract stroke of each online handwritten letter and then divided each stroke into some parts, i.e. tokens. The minimal features set are collected from these tokens and encoding to a classifier. The neural network classifier is designed with a robust weight initialization method. Finally, a database set of the Persian handwritten character samples has been employed to test the system in all their different shapes.

Fingerprint verification using statistical descriptors

The importance of high precision matching in fingerprint cannot be over-emphasized. This paper presents a novel fingerprint verification algorithm which improves matching accuracy by overcoming the shortcomings of poor image quality. The proposed method involves determination of a singular point using orientation field reliability, extraction of a square-sub-image (SSI); 129x129 pixels, statistical analysis of the co-occurrence matrices as well as application of dual analyses on experimental results; Pattern Recognition and Image Processing Laboratory (FVC2002) testing protocol and Program for Rate Estimation and Statistical Summaries (PRESS). The efficiency of the proposed method has been demonstrated by the experimental results which show equal error rate (EER) of 28% and a comparatively more accurate and robust means for reliable fingerprint verification.

Arabic Handwriting Recognition Using Projection Profile and Genetic Approach

This paper presents a complete system to recognize off-line Arabic handwriting. The proposed system starts from preprocessing and segmentation phases. It also includes thinning phase and finds vertical and horizontal projection profiles. The recognition phase is managed by genetic algorithm. The genetic algorithm stands on feature extraction algorithm that defines six features for each segment.

Self Organizing Features Map with Improved Segmentation to Identify Touching of Adjacent Characters in Handwritten Words

This paper presents an intelligent technique for segmentation of off-line cursive handwritten words particularly on touching characters problem. In this study, Self Organizing Feature Maps (SOM) is implemented to identify the touching portion of the cursive words. The image of the connected characters is preprocessed and the core-zone is detected to overcome ascender and descender of the touched character. Prior to clustering, the pixels of the image are mapped into coordinate system as features vector. These features vector are clustered into three classes: left, right and middle region, and the vertical segmentation is performed using SOM to determine the winner node of middle region.The experiments are conducted using syntactic CCC database.The results show that the proposed algorithm yields promising segmentation output and feasible with other existing techniques.

A Review on Fingerprint Classification Techniques

Automatic fingerprint recognition has been widely applied as a personal identification tool and biometrics applications due to their reliability and uniqueness features. One of the important tasks for any large scale fingerprint recognition system is fingerprint classification. Classifying a fingerprint images is a very difficult pattern recognition problem, due to the small interclass variability, the large intraclass variability, the presence of noise and the ambiguous properties of fingerprints. An accurate classification algorithm can greatly reduce the number of comparisons during fingerprint retrieval and consequently speeded up the identification process. Over the past few decades, a significant amount of researches and techniques has been proposed for distinguishing fingerprint classes. In this paper, we review existing approaches that have been applied to the classification problems. The explanation and discussion in this paper will cover the issues, designs and performance of several techniques for fingerprint classification system.