Chiun-Li Chin

DETECTION AND COMPENSATION ALGORITHM FOR BACKLIGHT IMAGES WITH FUZZY LOGIC AND ADAPTIVE COMPENSATION CURVE

This paper presents a new algorithm for detection and compensation of backlight images. The proposed technique attacks the weakness of the conventional backlight image processing methods such as over-saturation, losing contrast and so on. The proposed algorithm consists of two operation phases: detection and compensation phases. In the detection phase, we use the spatial position characteristic and histogram of backlight image to obtain two image indices, which can determine the backlight degree of an image. Fuzzy logic is then used to integrate these two indices into a final backlight index determining the final backlight degree of an image precisely. Second, in the compensation phase, to solve the over-saturation problem that exists usually in conventional image compensation methods, we propose the adaptive compensation-curve scheme to compensate and enhance the brightness of backlight images. The luminance of a backlight image is adjusted according to the compensation curve, which is adapted dynamically according to the backlight degree indicated by the backlight index estimated in the detection phase. The performance of the proposed technique is tested on 100 backlight images covering various kinds of backlight conditions and degrees. The experimental and comparison results clearly show the superiority of the proposed technique.

Using Fuzzy Inference and Cubic Curve to Detect and Compensate Backlight Image

This paper proposes a new algorithm method for detection and compensation of backlight images. The proposed technique attacks the weaknesses of conventional backlight image processing methods, such as over-saturation and diminished contrast. This proposed algorithm consists of two operational phases, the detection phase and the compensation phase. In the detection phase, we use the spatial position characteristic and the histogram of backlight images to obtain two image indices which can determine the backlight degree of an image. The fuzzy inference method is then used to integrate these two indices into a final backlight index which determines the final backlight degree of an image more precisely. The compensation phase is used to solve the over-saturation problem which usually exists in conventional image compensation methods. In this phase, we propose the adaptive cubic curve method to compensate and enhance the brightness of backlight images. The luminance of a backlight image is adjusted according to the cubic curve equation which adapts dynamically according to the backlight degree indicated by the backlight index estimated in the detection phase. The performance of the proposed technique was tested against 300 backlight images covering a variety of backlight conditions and degrees. A comparison of the results of previous experiments clearly shows the superiority of our proposed technique in solving over-saturation and backlight detection problems.

Ischemic stroke detection system with a computer-aided diagnostic ability using an unsupervised feature perception enhancement method

We propose an ischemic stroke detection system with a computer-aided diagnostic ability using a four-step unsupervised feature perception enhancement method. In the first step, known as preprocessing, we use a cubic curve contrast enhancement method to enhance image contrast. In the second step, we use a series of methods to extract the brain tissue image area identified during preprocessing. To detect abnormal regions in the brain images, we propose using an unsupervised region growing algorithm to segment the brain tissue area. The brain is centered on a horizontal line and the white matter of the brains inner ring is split into eight regions. In the third step, we use a coinciding regional location method to find the hybrid area of locations where a stroke may have occurred in each cerebral hemisphere. Finally, we make corrections and mark the stroke area with red color. In the experiment, we tested the system on 90 computed tomography (CT) images from 26 patients, and, with the assistance of two radiologists, we proved that our proposed system has computer-aided diagnostic capabilities. Our results show an increased stroke diagnosis sensitivity of 83% in comparison to 31% when radiologists use conventional diagnostic images.

The emotion recognition system with Heart Rate Variability and facial image features

Amid the rapid advance of technologies in recent years, people live at fast paces and suffer invisible pressures in family, at workplace as well as in school. Emotional undulation tends easily to lead to melancholy. Statistically, the number of bipolar disorder sufferers in the last three years was 40 times that in the past 10 years. What should be more attended to is that 90 percent of suicide cases involved a history of “depression” or “bipolar disorder”. Currently, to assess the mental pressure on an individual, questionnaires or interviews are applied and determination is made by physicians personal experiences. The development of a more viable quantitative method for assessing individual psychological pressure would facilitate self-management of health. This paper thus aims to CCD-capture human face, measure users physiological data using BtECG and determine his/her emotional status by using the fuzzy theory based on the analysis of human face, eyes and Heart Rate Variability. The features of measurements and the determination are also fed to database, which can serve as reference for doctors in consulting related psychological disorders in future.

Based on Fuzzy Linear Discriminant Analysis for Breast Cancer Mammography Analysis

According to a research report by the Department of Health, Executive Yuan, R.O.C. (Taiwan), breast cancer is the most common type of cancer in women, while the mortality rate of breast cancer of females over 40 years old is extremely high. If detected early, it can be treated early, and the mortality rate of breast cancer can be reduced. Therefore, the image processing technology has been adopted to automatically breast images, select suspicious regions, and provide alerts to assist in doctors¡¦ diagnosis, reduce misdiagnosis rates due to fatigue of doctors, and improve diagnostic accuracy. In order to assist physicians in clinical diagnosis, a set of breast cancer detection algorithm was designed in this paper through the Fuzzy theory and linear discriminant analysis(LDA). First, the images were automatically segmented, and then targeting this feature, the brightness values of suspicious regions were retained while the brightness of other areas was reduced to find the suspected location of the cancer. After that, with the images obtained, Law¡¦s Mask and grayscale value momentum-intensive technologies were adopted to find the texture of each area. Finally, the Fuzzy LDA was adopted to identify the texture of the cancers in order to capture images of the cancer areas. The experimental results show that the accuracy of the current detection methods can be improved to generate a breast cancer detection system used in the auxiliary medical diagnosis system, effectively reduce physicians¡¦ determination time, and improve accuracy.

A novel architecture for converting single 2D image into 3D effect image

A novel 2D to 3D effect image conversion architecture integrated image segmentation system and depth estimation is presented. Its objective is to describe the technique used to generate stereo pair images starting from a single image source (a view-point) and its related depth map. The conjunction between segmented image and depth map allows reconstructing artificially the binocular view producing a 3D effect depending on screen dimension and image resolution. The system consists of an online ICA mixture model that performs image segmentation and depth estimation method to obtain the depth information of image. The proposed system is capable of performing automatic multilevel segmentation of images, based on depth information obtained by the image depth estimation method. The system does not reconstruct the real 3D coordinates of any object inside the scene, but simply assigns the most comfortable shift to the source points to give a 3D-entertainment to the viewer. A simple smoothing filtering technique adaptively resolves the occlusions generated by shifting of the source points without introducing visible and/or annoying artifacts.

Increasing Visual Perception Brain Stroke Detection System

Stroke has been the third among the top ten leading causes of death in Taiwan. The danger is not only high mortality, but also gave the family a heavy economic. Recently, inspects the ischemic stroke the method to be divided into two categories: Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). The former CT is the most regularly adoption because done once the time of the latter MRI is longer than the time of the former CT, and the latter MRI is the most expensive than the former CT. Due to academic studies of ischemic stroke detection is very finite, it is low successful rate to correctly detect stroke initial period. In tradition, we have to depend on radiologists professional knowledge to diagnose ischemic stroke image, it will make radiologists increase fatigue so that made an error diagnostic. Therefore, we propose a Increasing visual perception brain stroke detection system. We use mathematic morphology to extract brain area. Then using median filter to remove noise, and uses canny edge detection to find out the edge of the brain tissue, and setting peak value in edge histogram as seed to perform region growing. Finally, we can clearly recognize the area of stroke. In the experimental result the recognition successful rate can reach 85%. Our successful rate cannot reach 100% because each patients brain CT image is not certainly upright and foursquare. Future research direction, we will use other methods to correct gradient and CT image for each patient to improve successful rate of detection.

INTELLIGENT PULMONARY EMBOLISM DETECTION SYSTEM

Pulmonary embolism (PE) is a blockage of the main artery of the lung or one of its branches by a substance that has travelled from elsewhere in the body through the bloodstream. Most of the traditional PE detection methods depend on the professional physicians judgment. Serious PE will lead to death. Therefore, diagnosis of PE is very important. In this paper, we develop an automatic PE detection system for relieving doctors load. It is divided into five parts — preprocessing, finding pulmonary, vessel searching, vessel tracking and evaluation. In the finding pulmonary part, we use active contour model (ACM) to extract pulmonary area. Next, we use cubic curve contrast enhancement method to enhance the contrast of branch vessel image. Its objective is to highlight the embolism area in the branch vessel. And, in the main vessel searching, we employ k-means algorithm and Gaussian mixture model (GMM) to find the main vessel part. Moreover, we propose an effective vessel tracking method to achieve vessel tracking. Finally, we invite three radiologists to help us evaluate the performance of our proposed system and to obtain the ground truth image of our system. We can calculate the precision rate of our system according to these ground truth images. The experimental results show that a total of 62 false positives were obtained for the 16 cases. It means that the ratio of FP/DS is 3.875. At 3.875 FP/DS, the classification successful rate of our system is about 83% in main vessel detection and 82.6% precision rate in branch vessel detection. Finally, our system can relieve doctors load according to the result of questionnaire analysis.

DEGENERATIVE DISC SEGMENTATION AND DIAGNOSIS TECHNOLOGY USING IMPORTANT FEATURES FROM MRI OF SPINE IN IMAGES

In this paper, we focus on the medical imaging segmentation techniques which are used in the study of spine diseases. In the medical reports, it is shown that common people worry more about the spine diseases caused by the disc degeneration. Because of the complex composition of the spine, which includes the spine bones, cartilage, fat, water and soft tissue, it is hard to correctly and easily find out the position of each cartilage in the spine images. This above problem always causes over-segmentation or unability to extract the cartilages. Thus, we propose an accurate and automated method to detect the abnormal disc. We combine two standard models with the threshold value to accurately identify the cartilage. Among the processing, we also solve the noising problems of spine image through morphological methods, removing the noncartilage areas using our proposed method, and find out the average height of the cartilages. Therefore, we can easily determine whether the disc is degenerated or not. In the experimental result, the segmentation accuracy of the extracted region by the proposed approach is evaluated by two criterions. The first criterion is statistical evaluation indices of image segmentation. It is evaluated by professional physicians manual segmentation, and the results show that our proposed method is easily implemented and has high accuracy, with the highest rate reaching 99.88%. The second criterion is a comparison evaluation index evaluated by our proposed system and other existence system. From this result, we know that our proposed system is better than other existence system.

The benign and Malignant Recognition System of Nasopharynx in MRI image with Neural-Fuzzy based Adaboost classifier

The current state of diagnosis of nasopharyngeal approach to image interpretation is artificial, but this way will be sentenced to cross-cultural experience for each practice discrimination and may lead to refractory time delay. We used the image taken from radiologists and applied the Unsharp Mask to enhance the image edge. In order to reduce the computation time and increase the efficiency, the doctor may specify an ellipse region of interest (ROIs) of nasopharynx in the image. After that, we use histogram equalization to denoise and use Otsu methods to obtain the threshold value to binaries it in this region. After performing the above two methods, we can successfully segment the tumor region of the nasopharynx. Then, we use texture and geometric feature extraction method to extract the tumor region feature, and trained the data by Neural-Fuzzy based AdaBoost classifier to recognize benign or malignant tumors of the nasopharynx. This paper hope that it can improve nasopharyngeal malignant cancer identification accuracy, and assist doctors make more accurate diagnosis through using tumor texture, hypertrophy and symmetrical distribution features in Neural-Fuzzy based Adaboost classifier.