D.N.F. Awang Iskandar

Social Media Retrieval Using Image Features and Structured Text

Use of XML offers a structured approach for representing information while maintaining separation of form and content. XML information retrieval is different from standard text retrieval in two aspects: the XML structure may be of interest as part of the query; and the information does not have to be text. In this paper, we describe an investigation of approaches to retrieve text and images from a large collection of XML documents, performed in the course of our participation in the INEX 2006 Ad Hoc and Multimedia tracks. We evaluate three information retrieval similarity measures: Pivoted Cosine, Okapi BM25 and Dirichlet. We show that on the INEX 2006 Ad Hoc queries Okapi BM25 is the most effective among the three similarity measures used for retrieving text only, while Dirichlet is more suitable when retrieving heterogeneous (text and image) data.

Combining image and structured text retrieval

Two common approaches in retrieving images from a collection are retrieval by text keywords and retrieval by visual content. However, it is widely recognised that it is impossible for keywords alone to fully describe visual content. This paper reports on the participation of the RMIT University group in the INEX 2005 multimedia track, where we investigated our approach of combining evidence from a content-oriented XML retrieval system and a content-based image retrieval system using a linear combination of evidence. Our approach yielded the best overall result for the INEX 2005 Multimedia track using the standard evaluation measures. We have extended our work by varying the parameter for the linear combination of evidence, and we have also examined the performance of runs submitted by participants by using the newly proposed HiXEval evaluation metric. We show that using CBIR in conjunction with text search leads to better retrieval performance.

Analysis of Packets Abnormalities in Wireless Sensor Network

Wireless Sensor Networks (WSNs) use tiny, inexpensive sensor nodes with several distinguishing characteristics: they have very low processing power and radio ranges, permit very low energy consumption and perform limited and specific monitoring and sensing functions. However, its security becomes an issue because in WSNs, there is virtual communication by passing the data through sensor via internet. Caused of its limited capability, an intruder can attack the communication easier. Furthermore, routing in wireless sensor networks has, to some extent, been reasonably well studied. However, most current research has focused primarily on providing the most energy efficient routing. There is a great need for both secure and energy efficient routing protocols in WSN. Therefore, this project studies about the packets in WSN. To achieve the objectives, this project used AODV routing protocol to analyze the packets abnormalities in WSNs by using simulation technique. To show the differentiations of packets behaviors, the simulations have been conducted on AODV routing protocol under malicious node and without malicious node. It also conducts an analysis of packets behavior on flooding attack.

Towards the Semantic Representation of Biological Images: From Pixels to Regions

Biomedical images and models contain vast amounts of information. Regrettably, much of this information is only accessible by domain experts. This paper describes a biological use case in which this situation occurs. Motivation is given for describing images, from this use case, semantically. Furthermore, links are provided to the medical domain, demonstrating the transferability of this work. Subsequently, it is shown that a semantic representation in which every pixel is featured is needlessly expensive. This motivates the discussion of more abstract renditions, which are dealt with next. As part of this, the paper discusses the suitability of existing technologies. In particular, Region Connection Calculus and one implementation of the W3C Geospatial Vocabulary are considered. It transpires that the abstract representations provide a basic description that enables the user to perform a subset of the desired queries. However, a more complex depiction is required for this use case.

iPepper: Intelligent pepper grading and quality assurance system

Pepper is a key export of the state of Sarawak (Malaysian Borneo); it produces 98% of Malaysias pepper. At present, processed pepper berries are graded manually. This process is time consuming and error prone as it is very much dependent on the experience of the pepper grader. To overcome these weaknesses, we propose a Pepper Grading System which employs image processing and machine learning approaches using image features and moisture content data of the pepper berries. For instance, from initial tests, a high correlation between the grade of pepper berries to the colour features has been detected. Using existing machine learning algorithms in WEKA, we have obtained a 100% accuracy in categorising the pepper berries into the correct grades. In addition, moisture content and colourometer readings provide another 2 other parameters which may complement the image features in accurately classifying the berries into the right grades.

Multiclass brain Glioma tumor classification using block-based 3D Wavelet features of MR images

With the advent of more powerful computing devices, system automation plays a pivotal role. In the medical industry, automated image classification and segmentation is an important task for decision making about a particular disease. In this research, a new technique is presented for classification and segmentation of low-grade and high-grade glioma tumors in Multimodal Magnetic Resonance (MR) images. In the proposed system, each multimodal MR image is divided into small blocks and features of each block are extracted using three Dimensional Discrete Wavelet Transform (3D DWT). Random Forest classifier is used for the classification of multiple Glioma tumor classes, then segmentation is performed by reconstructing the MR image based on the classified blocks. MICCAI BraTS dataset is used for testing the proposed technique and experiments are performed for Low Grade Glioma (LGG) and High Grade Glioma (HGG) datasets. The results are compared with different classifiers e.g. multilayer perceptron, radial basis function, Naïve Bayes, etc., After careful analysis, Random Forest classifier provided better precision by securing average accuracy of 89.75% and 86.87% is obtained for HGG and LGG respectively.

Spatio-temporal semantic representation of Cardiac MRI in heart attack patients

Semantic Web technologies, applications and tools have made great steps forward in the life science and health care data exchange. However, developing appropriate semantic representations, including designing spatio-temporal ontologies, remains difficult and challenging. In this paper, we describe a framework to engineer a spatio-temporal semantic representation for the Cardiac MRI images using the current existing case studies conducted in Sarawak General Hospital Heart Centre.

Segmentation of Melanoma Skin Lesions Using Anisotropic Diffusion and Adaptive Thresholding

Segmentation is the first and most important task in the diagnosis of skin cancer using computer-aided systems and due to complex structure of skin lesions, the automated process may lead to a completely different diagnosis. In this paper, a novel segmentation method of skin lesions is proposed which is both effective and simple to implement. Smoothing of skin lesions in original image plays a pivotal role to secure an accurate segmented image. Anisotropic Diffusion Filter (ADF) is used in the initial stage to smooth images with preserved edges. Adaptive thresholding is then applied to segment the skin lesion of the image by binarizing it. The morphological operations are applied for further enhancement and final segmented image is obtained by applying proposed boundary conditions in which objects are selected on basis of distance. The proposed technique is tested on over 300 images and averaged results are compared with existing methods like L-SRM, Otsu-R, Otsu-RGB and TDLS. The proposed method achieved an average accuracy of 96.6%. Visual results for selected images also depicted better performance of proposed method even in the presence of bad illumination and rough skin lesions in the image.

Automatic Segmentation of CMRIs for LV Contour Detection

Research on detecting, recognising and interpreting cardiovascular magnetic resonance images (CMRIs) has started since the 1980s. Time consuming and the need of expert evaluation are the key problems in the manual tracing efforts of CMRIs in a routine investigation. CMRIs manual tracing is also dependent on image quality, and there is no one-size-fits-all MRI setting for an optimum image result. In this paper, we present an approach using 2-Standard Division (2-SD) correlation along with the Sum of Absolute Difference technique and Otsu Watershed to automatically detect the left ventricle (LV) wall and blood pool in the effort to automatically assist the assessment of cardiac function. We test the approach using the Sunnybrook Cardiac Data, a standard benchmark dataset. The results shown that the proposed method had improved the automatic detection of the epicardium and endocardium.

Automated machine for sorting Sarawak pepper berries

White pepper berries is one of the Malaysias key export as it is categorised as high valued commodity product. At present, processed white pepper berries are graded semi-automatically. This process is time consuming as it dependent on the experience of the pepper grader. In this paper we present a solution for Sarawak White Pepper grading using a combination of image processing technique and robotic solutions to sort pepper berries into their respective grades. In particular, we present the result of using different colour sensors. With the automated sorting machine, more high grades pepper berries are able to be sorted; this means more income to the smallholders, which are the local pepper farmers.