Sarina Mansor

A content based image retrieval system for marine life images

Malaysia has been recognized as one of the twelve nations endowed with rich biodiversity. Such huge number of species in the rain forest and sea are an important asset that need to be properly documented. Responding to these important needs, we have designed and evaluated a content based image retrieval system catered for marine life images. This paper investigates the effectiveness of various low level image descriptors, which includes the colour, shape and texture features in representing the semantic categories of the marine life images. This is a challenging task since the images are taken from different viewpoints and marine species do not possess consistent colour, shape and textural appearance. For the purpose of evaluating the overall image retrieval system effectiveness, we design and implement an image retrieval system which support image query by example. Experiment is conducted to evaluate the effectiveness of various low level image descriptors and the fusion of multiple features. The experiment result on the image retrieval performance is presented.

Lossless compression of Fluoroscopy medical images using correlation and the combination of Run-length and Huffman coding

Medical centers produce a massive amount of sequential medical images for examinations such as CT, MRI and Fluoroscopy, where each examination of a patient consists of a series of images. This takes up a large amount of storage space, in addition to the cost and time incurred during transmission. For medical data, lossless compression is preferred to the greater gains of lossy compression, in the interest of accuracy. This paper proposes a new method for lossless compression of pharynx and esophagus fluoroscopy images, using correlation and combination of Run Length and Huffman coding on the difference pairs of images classified by correlation. From the experimental results obtained, the proposed method achieved improved performance with a compression ratio of 11.41 for the proposed combination of Run-length and Huffman coding (RLHM-D) on the difference images as compared to 1.31 for the standard images.

Segmentation and compression of pharynx and esophagus fluoroscopic images

Enormous amounts of sequential medical images are produced in modern medical examinations, typically in Fluoroscopy. Although highly effective, such large quantities of images incur a high cost in terms of storage, processing time and transmission. This paper proposes a method for lossless compression of targeted parts within Fluoroscopy images, extracting the region of interest (ROI) - in this case the pharynx and esophagus, and employing customized correlation and the combination of Run Length and Huffman coding, to increase compression efficiency. The experimental results show that the proposed method improved performance with a compression ratio of 300% better than conventional methods.

Lossless Compression of Pharynx and Esophagus in Fluoroscopic Medical Images

 Abstract—Hospitals and medical centers produce a tremendous amount of sequential images for medical examinations such as MRI, CT and Fluoroscopy. This series of images takes up a large amount of storage space, in addition to the cost and time incurred during transmission. For medical data, lossless compression is preferable to the greater gains of lossy compression, in the interest of reliability. This paper proposes a new method for lossless compression of pharynx and esophagus fluoroscopy images, depending on correlation and combination of Run Length and Huffman. Otherwise, the shifted images moved to a shifted group and compress separately. From the experimental results obtained, the proposed method achieved improved performance with a compression ratio of 12.2 for the proposed combination of Run-length and Huffman coding (R. Huff) on the difference images as compared to 1.35 for the standard method.

Content-Based Image Retrieval system for marine life images using gradient vector flow

Content Based Image Retrieval (CBIR) has been an active and fast growing research area in both image processing and data mining. Malaysia has been recognized with a rich marine ecosystem. Challenges of these images are low resolution, translation, and transformation invariant. In this paper, we have designed an automated CBIR system to characterize the species for future research. Gradient vector flow (GVF) has been implemented in a lot of image processing applications. Inspired by its fast image restoration algorithms we applied GVF for marine images. We evaluated different automated segmentation techniques and found GVF showing better retrieval results compared to other automated segmentation techniques.

An improved retrieval performance with hybrid shape descriptor and feature matching

Research on Content Based Image Retrieval (CBIR) has become popular as it offers solutions to overcome or complement the drawbacks of Text Based Image Retrieval (TBIR). In CBIR, feature extraction and feature matching are two critical processes, which are of high importance to the retrieval performance of the system. This paper introduces a new approach to shape-based image retrieval by combining global and local shape features using Zernike moments (ZM) and edge-gradient co-occurrence matrix (EGCM) respectively. Two-stage matching strategy is then used to measure similarity between images. Our proposed method achieves higher precision rate compared to other commonly used shape feature.

Classification of benign and malignant tumors in histopathology images

Breast cancer leads the list of cancer that act on women worldwide. It starts when cells in the breast begin to build up beyond control. These cells normally create a tumour that can usually be seen on an x-ray or felt as a lump. Analysing and grading the tumour will take up much of a pathologist time. Pathologists have been largely diagnosing disease the same way for the past years, by manually reviewing images under a microscope. Thus, to help the pathologists improve accuracy and significantly change the way breast cancer been diagnosed, this paper presents an automated classification program. BreakHis dataset was used which build of 7909 breast tumor images gathered from 82 patients. This system is developed in order to categorize the cancer cells into two classes of cancer which are benign and malignant. The classification system compared different types of feature extractors using k-nearest neighbours classifier to efficiently observe the performance of the classification system. An extensive set of experiments showed that the overall accuracy rates range from 83% to 86%.

Improving shape descriptor complexity via wavelet decomposition

Research on content based image retrieval (CBIR) has received a considerable attention as it offers solutions to overcome and complement the drawbacks of text based image retrieval (TBIR). One of the crucial studies in this system is the feature extraction process where the low level features, i.e. shape, color and texture are the common features used to describe the image content. However, studies in the past only focus on deriving good descriptors from these low level features and less attention has been given on the complexity improvement of these descriptors. This paper proposes a simple technique to reduce the complexity computations of shape feature via decomposition method. We employ the discrete wavelet transform as the decomposition technique and use the transform image content to derive the shape feature. Our method has shown an improvement of speed performance of more than 50% compared to the conventional method. The database used in this study is the MPEG7 database consisting of 1400 images with 70 classes.