Information Processing And Optimization Of Artificial Vision For Retinal Prosthesis

Blindness is one of the most serious disabilit ies. Retinal prosthesis is proposed as a way of restoring vis ion for such a blinding disease as age-related macular degeneration and retinitis pigmentosa by electrically stimulating remaining retina, which presents the “artificial vision” to the subject in the form of discrete phosphene. However, due to the challenges such as electrode fabrication, power consumption and long-term viability, the current number of implanted electrodes is very limited, thus leads to is a low-resolution representation of prosthetic vision. Meanwhile, clinical tests have reported that the brightness or color information, or depth perception by current prosthetic vision are far from normal vis ion. The capacity and performance of the prosthetic device should be significantly improved for practical use. At present, technological advancements in retina prosthesis, coupled with recent scientific investigations, have transformed the focus of the field from creating visual sensations through electrical stimulation to optimizing the perception. Focusing on this issue and using information science and cognitive science, we put forward image strategies for information processing and optimization in artificial prosthetic vis ion. This thesis consisted of the following parts.To solve the problem of the low-resolution, a lack in color and depth information, and limited dynamic range of brightness in the prosthetic vis ion, we propose object extraction and optimization strategies based on a visual saliency model to enhance the ability of object search and recognition under prosthetic vision. and we investigated the effect of the strategies under simulated prosthetic vision. Results showed that saliency-based image processing could significantly increase recognition accuracy nearly for two fold. The saliency-based image processing strategies were verified beneficial to recognition performance of object in scenes under prosthetic vision.Face recognition is an important cognitive activity in daily life. Optimizing the face recognition under prosthetic vision would help improv prosthetic wearer’s quality of life. Based on the face detection technique, we studied the face region detection and optimization for enhancing face identification in natural scene. Results showed that human face can be effectively located and detected in the low resolution prosthetic vis ion under the confined vis ion field. Meanwhile, recognition could be significantly improved for familiar faces. The face-detection-based method offers new strategies for face location and recognition under limited prosthetic vis ion. Additionally, results revealed that certain face external characteristics can facilitate recognition performance in prosthetic vision.Moving object recognition is of great importance for navigating task in living environment for prosthesis wearers. Using background subtraction, we proposed to optimize the moving object recognition strategies, which was tested in simulated experiments. By introducing an universal background subtraction technique(ViBe), the moving object was effectively segmented in daily-life scenes. The experimental results showed that by two kinds of image optimization(background reduction and background reduction & foreground enhancement), the indoor and outdoor recognition accuracy under 24 × 24 resolution significantly increased from 49.06% and 44.06% to 78.44% and 85.31% respectively. When the resolution increased to 32 × 32, the highest recognition accuracy could reach 89.38%. By background-subtraction-based strategies, the recognition of moving objects in daily scenes could be markedly enhanced. Also, resolution of the prosthetic vis ion affected the recognition performance significantly. These strategies will assist implant recipients to avoid dangerous situations and attain independent mobility in daily life.These image-based visual information processing will help to optimize the design of retinal prosthesis, which provides solid experimental and theoretical basis to visual function rehabilitation for the blind.