Development of Novel Image Enhancement Methods for Bio-medical Images

Abstract

Image enhancement is a critical image processing approach that emphasizes critical information in an image and minimizes or eliminates some secondary information in improving the identification quality. The purpose is to make the objective images more application-specific than the original images. The primary goal of image enhancement is to improve the interpretability and visual quality of images, aiming to enhance the perception of information contained within the pictures for human viewers. Additionally, it can serve as a valuable pre-processing step for machine learning algorithms. The concept of image enhancement has been extensively employed in various fields of study, including biology, computer vision, remote sensing, and medical imaging. Physicians often utilize medical images for illness diagnosis, disease monitoring, and therapy planning. Medical imaging methods can produce detailed human body images in vivo. Organs and tissue are shown structurally and functionally via the pictures obtained. This information may be utilized to aid in diagnosing and treating patients. Among all radiology examinations, the computed tomography (CT) scan is the most popular and frequently utilized by medical professionals since it is less expensive and reveals comprehensive structures of inside body components. Histopathology is another critical tool for physicians to determine the causes and consequences of all illnesses and disorders. The fundus image is the most often acquired image for diagnosing different eye diseases. This thesis's most often used medical images are CT scans, digital histopathology images, and retinal fundus images. CT, Histopathology, and Fundus pictures all exhibit a variety of image abnormalities, including low contrast, noise, blur, and distorted features, which vary according to the kind of medical imaging modality used. Medical image enhancement is a key technique for improving diagnostic images for illness diagnosis, disease monitoring, and treatment planning. Although existing image enhancement methods are useful for a variety of application areas, the effects are often inappropriate for medical images due to the presence of multiple defects, color distortion, noise amplification, under or over enhancement, brightness degradation, and information distortion. In order to address these shortcomings, this dissertation vii proposes a variety of medical image enhancement algorithms for CT scans, digital histopathology images, and retinal fundus images. The first algorithm is based on optimal morphological transforms for improving the noise and contrast of CT images in the wavelet domain. Initially, the optimal morphology transform (OMT) technique is employed to increase the contrast of the original CT image, and then the discrete wavelet transform (DWT) method is utilized to partition the original CT image and the OMT transformed image into four separate portions. For increased contrast while preserving image brightness, the singular value decomposition (SVD) technique is utilized on LL sub-band output alone, while the Edge Map (EM) method is applied on the HH sub-band of the original CT image to denoise it. The proposed method outperforms the other six enhancement approaches. The experimental findings indicated that the suggested approach is capable of properly enhancing CT images while preserving their inherent features. The proposed approach outperforms existing methods such as S-curve [85], AGC-DWT [69], AGC-WHD [65], GAGC-DWT [61], EBCHE [50], and TCHE-DWT [51] in terms of EME, CII, Michelson contrast ratio and Weber contrast ratio, peak signal to noise ratio (PSNR), and entropy of the enhanced images. The proposed technique is appropriate for enhancing CT images and may be used to help pathologists or physicians in making the correct diagnosis. The second algorithm effectively enhances non-contrast CT images with dual tree complex wavelet transform (DT-CWT) and adaptable morphology. This new approach for effectively enhancing non-contrast CT images is introduced, which employs a dual tree complex wavelet transform (DT-CWT) technique and adaptable morphology for efficient high-frequency sub-bands denoising and low-frequency sub-bands enhancement. Experiments on structured CT scan images were conducted to evaluate the proposed procedure's success on quantitative and subjective tests. Furthermore, experiments on CT image databases explicitly show that the proposed methodology outperforms other conventional enhancement techniques to improve and maintain fine information devoid of noise. The suggested algorithm produces greater brightness preserved CT images with higher contrast and lower noise levels, with all regions becoming visible and prominent. The suggested approach is appropriate for CT image enhancement and may be used to help pathologists or physicians precisely diagnose the disease. The third algorithm seeks to improve the color histopathology image contrast based on viii retinex theory and local contrast adjustment. To begin, a novel multiscale retinex with adaptive weighting is presented to enhance the contrast of a color histopathological image in the HSV color model, and then the image's local details are strengthened using a new weighted Contrast Limited Adaptive Histogram Equalization (CLAHE) methodology applied to the luminosity component in the L*a*b* color model. Experiments using histopathology images have been used to assess the proposed procedure's success on both quantitative and subjective measures. Additionally, experiments conducted using histopathology image databases reveal strongly that the proposed methodology outperforms existing standard enhancement strategies in terms of overall contrast improvement and fine detail preservation. The proposed technique generates brightness preserved, better contrast natural histopathological images with no image artifacts, with all areas being apparent and noticeable. The proposed methodology is appropriate for histopathology image enhancement and may be utilized to support pathologists or doctors in making accurate diagnoses. The fourth algorithm is an effective image enhancement method for improving the luminosity and contrast of color retinal fundus images. The purpose of this algorithm is to provide an efficient technique for color retinal fundus image enhancement that is based on luminance and contrast improvement. To begin, the luminosity of the color retinal image is improved using a novel just noticeable difference (JND)-based adaptive gamma correction method, followed by multiple layers of CLAHE in the L*a*b* color mode to enhance retinal image contrast. Experiments using benchmark test retinal images are utilized to evaluate the proposed method's performance on qualitative and quantitative metrics. Furthermore, findings from experiments on retinal image databases demonstrate vividly that the proposed approach surpasses other conventional enhancement approaches. The proposed approach produces natural-looking fundus images with increased contrast in which all areas are visible and distinct. The proposed technique is suitable for color retinal image enhancement, which can be used to help ophthalmologists or clinicians inaccurately diagnose disease. Each of the novel methods is presented, described thoroughly, and applied to a set of medical images from a publicly available database to measure its effectiveness qualitatively and quantitatively in terms of various performance parameters, such as signal-to-noise ratio, discrete entropy, edge preservation index, contrast ratio, and enhancement ratio. There are significant quantitative and qualitative findings that show improvements when compared to the reference enhancement methods. These results ix indicate that the newly developed algorithms represent a valuable contribution to advancement in this field.