Image Enhancement Methods for Anterior Segment Photographed Images

Abstract

In the digital age, the use of smartphone digital camera images as an imaging modality for early eye disease screening and detection has been widely explored. Nevertheless, the process of taking pictures often raises some issues that may be related to the dynamic range limitations of cameras and uneven lighting, which may cause uneven illumination in the images. Therefore, image enhancement is further needed to improve the visual quality of the images and to provide sufficient detail for basic computer vision tasks including segmentation. This work studies the effect of image enhancement methods on the segmentation results of anterior segment photographed image (ASPI) ocular regions with nonuniform illumination. During the pre-processing stage, the image enhancement methods are performed on the nonuniform illumination ASPI. Those images are later trained using a deep learning approach for semantic segmentation. Based on the network performance, the Multiscale Retinex with Chromaticity Preservation (MSRCP) method gives the best segmentation results for all classes with mean accuracy of 0.921, mean IoU of 0.805 and mean BFScore of 0.717. In addition, the image quality assessment is performed on the ASPI with image enhancement methods. It is found that the images using MSRCP method had the lowest values for Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), Naturalness Image Quality Evaluator (NIQE), and Perception-based Image Quality Evaluator (PIQE), indicating the highest image quality compared to the other methods. For future work, the segmented region of the eyes, including the sclera, iris, and pupil, can be utilized as a region of interest (ROI) to detect various anterior eye diseases.