Shyr-Shen Yu

A novel optic disc detection scheme on retinal images

Robust and effective optic disc detection is a necessary processing component in automatic retinal screening systems. In this paper, optic disc localization is achieved by a novel illumination correction operation, and contour segmentation is completed by a supervised gradient vector flow snake (SGVF snake) model. Conventional GVF snake is not sufficient to segment contour due to vessel occlusion and fuzzy disc boundaries. In view of this reason, the SGVF snake is extended in each time of deformation iteration, so that the contour points can be classified and updated according to their corresponding feature information. The classification relies on the feature vector extraction and the statistical information generated from training images. This approach is evaluated by means of two publicly available databases, Digital Retinal Images for Vessel Extraction (DRIVE) database and Structured Analysis of the Retina (STARE) database, of color retinal images. The experimental results show that the overall performance is with 95% correct optic disc localization from the two databases and 91% disc boundaries are correctly segmented by the SGVF snake algorithm.

Automatic detection of abnormal mammograms in mammographic images

We develop a computer aided system to detect abnormal mammograms.We extract only 5 features of intensity and gradient for mass detection.Principal component analysis is applied to determine the feature weights.The abnormality detection classifier by feature weight adjustments is proposed.We evaluate our method upon 2 different datasets. This paper proposes a detection method for abnormal mammograms in mammographic datasets based on the novel abnormality detection classifier (ADC) by extracting a few of discriminative features, first-order statistical intensities and gradients. As tumorous masses are often indistinguishable from the surrounding parenchyma, automatic mass detection on highly complex breast tissues has been a challenge. However, most tumor detection methods require extraction of a large number of textural features for further multiple computations. The study first investigates image preprocessing techniques for obtaining more accurate breast segmentation prior to mass detection, including global equalization transformation, denoising, binarization, breast orientation determination and the pectoral muscle suppression. After performing gray level quantization on the breast images segmented, the presented feature difference matrices could be created by five features extracted from a suspicious region of interest (ROI); subsequently, principal component analysis (PCA) is applied to aid the determination of feature weights. The experimental results show that applying the algorithm of ADC accompanied with the feature weight adjustments to detect abnormal mammograms has yielded prominent sensitivities of 88% and 86% on the two respective datasets. Comparing other automated mass detection systems, this study proposes a new method for fully developing a high-performance, computer-aided decision (CAD) system that can automatically detect abnormal mammograms in screening programs, especially when an entire database is tested.

An improved GVF snake based breast region extrapolation scheme for digital mammograms

To accurately extrapolate the breast region from a mammogram is a crucial stage of breast mass analysis. It significantly influences the overall analysis accuracy and processing speed of the whole breast mass analysis. In this paper, a novel edge map adjusting gradient vector flow snake (EMA GVF snake) algorithm for extrapolation of breast region from mammograms is proposed. In the proposed algorithm, the median filter is used to filter out the noise in a mammogram, the scale down stage is used to resize down the mammogram size (hence speeding up the extrapolation). The binarization processing stage and the morphological erosion processing stage are used to find a rough breast border. Then a novel gradient adjusting stage is applied to get a modified edge map and the gradient vector flow snake (GVF snake) is used to get the accurate breast border from the rough breast border. The proposed algorithm is tested on 322 digital mammograms from the Mammogram Image Analysis Society database. The mean error function, misclassification error function and the relative foreground area error function are conducted to evaluate the results of the detected breast border and the extracted breast region. Experimental results show that the breast border extrapolated by the proposed algorithm approximately follows the breast border extrapolated by an expert radiologist. Experimental results also show that the proposed algorithm is more robust and precise than the traditional GVF snake scheme for the breast extrapolation on mammograms.

Bone age cluster assessment and feature clustering analysis based on phalangeal image rough segmentation

There are different feature selections in a bone age assessment (BAA) system for various stages of skeletal development. For example, diameters of epiphysis and metaphysis are used as sensitive factors during the early stage. Once the epiphyseal fusion has started, an additional feature such as the degree of fusion is extracted at the later stage. Image analysis is a critical point for feature selections to get a fine BAA, which includes ROI processing and feature extraction. Nevertheless, the related modeling techniques are various depending on the characteristics of different stages of bone maturity, which usually are taken as a priori knowledge in most previously proposed schemes. If a coarse bone age cluster (stage) for a hand radiograph could be automatically pre-assigned, then these corresponding image analysis methods can be identified. This could avoid taking a priori knowledge and provide a more flexible and reliable BAA system. For this purpose, a bone age cluster assessment system using fuzzy neural network (FNN) based on phalangeal image rough segmentation is presented in this work. This system includes two parts. The first part adjusts the feature weights to stable conditions according to four new defined bone age stages, which satisfy feature development of epiphysis and metaphysis. The second part is bone age cluster assessment on hand radiography based on the results of the first part. Experimental results reveal that the presented FNN system provides a very good ability to assign a hand radiograph to an appropriate bone age cluster and demonstrates the rationality of those new defined stages. Furthermore, the related feature clustering analysis for various stages is discussed to provide an accurate quantitative evaluation of specific features for the final BAA.

Alpha-gamma equalization-enhanced hand radiographic image segmentation scheme

Bone age assessment of children is a procedure frequently performed in pediatric radiology. The feature extractions of metaphyseal and epiphyseal regions are crucial to automatic bone age assessment. The first step of feature extraction is applying a segmentation scheme to find exact regions of epiphysis and metaphysis. A segmentation method is normally based on both intensity information and the relative location of pixels. There is a fundamental problem; when the intensity contrast of soft tissue and bony tissue is poor, bony and soft tissue cannot easily be separated. We propose an α-gamma equalization method to increase the intensity contrast between bony and soft tissue. Sobel, two-means, Canny edge-detection, and watershed methods are applied to illustrate the effect of this method on initial segmentation. Adaptive two-means and gradient vector flow snake are adopted for the final segmentation. Experimental results reveal that α-gamma equalization-enhanced two-means initial segmentation with an adaptive two-means clustering scheme can greatly reduce the average error measurements of segmentations. These are evaluated according to the following measurements: misclassification error, edge mismatch, region nonuniformity, relative foreground area eror, and modified Hausdorff distance. Furthermore, the experimental results show that the proposed scheme provides increased stable performance for the segmentation of epiphyseal/metaphyseal regions.

General Regression Neural Network utilized for color transformation between images on RGB color space

The color transformation of images is utilized to enhance some potential features of a color image for wide applications in different fields. This paper presents a General regression analysis based algorithm for global color transformation between images. In this paper, General Regression Neural Network (GRNN) is respectively conducted on the R, G, and B planes of the target and source images to find the corresponding best fitting functions of each plane. The new values of each pixel of target image are then evaluated by using the best fitting functions. These new values are then combined into the result RGB color image. The experimental results show that the presented approach has two major advantages: (A) the proposed algorithm is manual free, simple, effective and accurate in transferring color between images without any change in the image details, (B) there are no restrictions in the image dynamic color ranges in the proposed algorithm.

A Novel Iris Segmentation Scheme

One of the key steps in the iris recognition system is the accurate iris segmentation from its surrounding noises including pupil, sclera, eyelashes, and eyebrows of a captured eye-image. This paper presents a novel iris segmentation scheme which utilizes the orientation matching transform to outline the outer and inner iris boundaries initially. It then employs Delogne-Kasa circle fitting (instead of the traditional Hough transform) to further eliminate the outlier points to extract a more precise iris area from an eye-image. In the extracted iris region, the proposed scheme further utilizes the differences in the intensity and positional characteristics of the iris, eyelid, and eyelashes to detect and delete these noises. The scheme is then applied on iris image database, UBIRIS.v1. The experimental results show that the presented scheme provides a more effective and efficient iris segmentation than other conventional methods.

A Fovea Localization Scheme Using Vessel Origin-Based Parabolic Model

At the center of the macula, fovea plays an important role in computer-aided diagnosis. To locate the fovea, this paper proposes a vessel origin (VO)-based parabolic model, which takes the VO as the vertex of the parabola-like vasculature. Image processing steps are applied to accurately locate the fovea on retinal images. Firstly, morphological gradient and the circular Hough transform are used to find the optic disc. The structure of the vessel is then segmented with the line detector. Based on the characteristics of the VO, four features of VO are extracted, following the Bayesian classification procedure. Once the VO is identified, the VO-based parabolic model will locate the fovea. To find the fittest parabola and the symmetry axis of the retinal vessel, an Shift and Rotation (SR)-Hough transform that combines the Hough transform with the shift and rotation of coordinates is presented. Two public databases of retinal images, DRIVE and STARE, are used to evaluate the proposed method. The experiment results show that the average Euclidean distances between the located fovea and the fovea marked by experts in two databases are 9.8 pixels and 30.7 pixels, respectively. The results are stronger than other methods and thus provide a better macular detection for further disease discovery.

A study of vessel origin detection on retinal images

Parabolic model is commonly used for fovea and macular detection. But, the center of an optic disc is mostly taken as the vertex of the parabola-like vasculature. Since vessels generate out from the vessel origin, taking vessel origin as the vertex can provide better fovea localization than taking optic disc center. Recently, the vessel origin is also used to detect vessels within an optic disc. However, there is no published research for finding the exact vessel origin position. This paper proposed a novel method to locate the position of vessel origin. First, a retinal image is processed to get the vascular structure. Then, four features based on the characteristic of vessel origin are selected, and Bayesian classifier is applied to locate the vessel origin. The proposed method is evaluated on the publicly available database, DRIVE. The experimental results show that the average Euclidean distance between the vessel origin and the one marked by experts are 13.3 pts, which are much better than other methods. This can further provide a more accurate vessel and fovea detection.

Reversible fragile watermarking scheme for three-dimensional models

Watermarking is a very important technique for protecting the authorization of digital content, such as still images, video streams, audio streams, and 3-D models. Most fragile watermarking schemes for 3-D models are not reversible. Published reversible or irreversible fragile watermarking schemes for 3-D models have many drawbacks, such as the causality problem, not being blind, or being unable to locate changed regions. To overcome these drawbacks, a reversible fragile watermarking scheme for 3-D models in spatial domain is proposed. Principal component analysis (PCA) is employed to produce the PCA coordinate system and make the system robust against similarity transformation attacks. PCA, together with a novel interval embedding technique, provide the blind reversibility of the proposed scheme. Intervals larger than a threshold and smaller than another threshold are adopted as legitimate to avoid large distortion. Moreover, experimental results show that the proposed scheme can overcome the causality, convergence, and confusion problems; provide both tampering detection and better embedding rate and requiring a small key size.