Super-resolution (SR) is an elegant technique that can reconstruct high-resolution (HR) videos/images from their lowresolution (LR) counterparts. Most of the conventional SR methods utilize linear mappings to learn complex LR-to-HR relationships, where these linear mappings are often learned from training. Inspired by our previous linear mapping based SR method [1], we propose a novel super-interpolation based SR method that utilizes adjusted self-exemplars. That is, in order to find sufficient amounts of LR-HR patch pairs in self-exemplars, we iteratively augment selfexemplars from an LR input image to create additional selfexemplars. In doing so, our proposed SR method is able to find welllearned linear mappings on-line from self-exemplars without using external training images, and outperforms other conventional SR methods.
Super-resolution (SR) image processing describes any technique by which the resolution of an imaging system is enhanced. Normally, the resolution being enhanced is spatial; images are processed to provide noise reduction, sub-pixel image localization, etc. Less often, it is used to enhance temporal properties – for example, to derive a higher framerate sequence from one or more lower framerate sequences. Time domain continuous imaging (TDCI) representations are inherently frameless, representing a time-varying scene as a compressed continuous waveform per pixel, but they still imply finite temporal resolution and accuracy. This paper explores computational methods by which the temporal resolution can be enhanced and temporal noise reduced using a TDCI representation.
To capture an image of object at long distance, viewing direction of the camera must be controlled when long-focal length is used. Therefore, there is method of control viewing direction at high-speed using galvanometer scanner. In this method, vibration of the mirror influences captured image due to mechanical motion. To get image of an object without mechanical motion, a camera should get an image of entire field of view. Then, object image can be got by clipping it from the image of entire field of view. In this method, camera must get also information other than the object at once to capture an entire image with wide range. Therefore, the object image is low-resolution because a number of image sensor's pixel which contributes the object is few. Thus, we propose a camera which has controllability of viewing direction and can capture highresolution image without mechanical motion based on computational imaging. By using this camera, it is possible to control at high-speed. Moreover the camera has no influence of vibration. We did a simulation and confirmed that the proposed camera which has controllability of viewing direction got highresolution image of an object without mechanical motion.
In general, edges in the peripheral areas of around view monitor (AVM) wide-angle (WA) images tend to be blurred. This paper proposes a self-example-based edge enhancement algorithm to improve the definition of such edges. First, a low-resolution (LR) version of a blurred WA high-resolution (HR) image is produced via down-scaling. Next, a proper self-example for each non-overlapped patch in the HR image is found within the LR image in terms of selfsimilarity. Then, high frequency information is extracted from the found LR patch, and it is finally added to the input HR patch. Experimental results show that the proposed algorithm provides higher JNBM values than previous works with outstanding visual quality.