Smooth downsampling of depth images for visual prostheses

Over the past few decades, a variety of visual prostheses is developed to allow for the restoration of the vision for the blind. In visual prostheses, visual perception is limited to extremely low image resolution mainly due to restrictions in the fabrication of efficient microelectrode arrays. As a result, tasks such as navigation and way finding become difficult for those using implantable visual prostheses. Depth cue is a suitable alternative to intensity images to improve the quality and success of the aforementioned tasks in patients. After the processing of depth images, intensity of an object depends on its distance from the patient. Based on this principle, a method for preprocessing and downsampling of the depth images is proposed in this paper. We propose a method to enhance the contrast of the depth images and downsample the results to 6 × 12 images. This paper analyzes common downsampling methods and proposes a method based on the mode function. In the proposed method, the mode function is applied on every four successive frames to use temporal information in addition to stationary information. Quantitative and qualitative evaluations upon the LIRIS dataset are presented to compare the results of proposed method with rivals.