Image Processing Projects

Abstract: Computer vision and computational photography rely on multispectral imaging. Image registration is needed to avoid spectral information distortion from imaging device movement. Multispectral data registration measures are robust but computationally intensive. SSD and SAD…

Abstract: Binocular stereo vision (SV) is widely used to reconstruct depth information, but strong occlusions make it vulnerable. Light-field (LF) imaging, an emerging computational photography technology, records multiple angular views in a single exposure to…

Abstract: Cataract is the leading cause of blindness worldwide. Accurate and convenient cataract detection and severity evaluation will help. This paper proposes automatic cataract detection and grading. With prior knowledge, improved Haar and visible structure…

Abstract: The guided filter and its derivatives are widely used in image processing and computer vision due to their low complexity and good edge-preservation. Despite this success, the guided filter variants cannot handle more aggressive…

Abstract: Marine engineering and aquatic robotics have made underwater image enhancement popular. Recently, many underwater image enhancement algorithms have been proposed. These algorithms are mostly tested on synthetic datasets or a few real-world images. Thus,…

Abstract: Edge detection is an essential preprocessing step for high-level tasks in image analysis and computer vision. Since image contents vary, it’s hard to set a universal threshold. This project introduces a real-time adaptive, robust,…

Abstract: Many practical applications requiring multiple related images require image registration. This project proposes a method to handle geometric and intensity transformations in image registration. Modifying an accurate and fast elastic registration algorithm (Local All-Pass-LAP)…

Abstract: Imaging arterial mechanical properties may help diagnose vascular disease. Pulse wave velocity (PWV) measures arterial stiffness and cardiovascular mortality. PWI is a high-resolution method for imaging pulse wave propagation. We present adaptive PWI, a…

Abstract: In single image dehazing, image noise causes depth-dependent visual artifacts. Most dehazing methods use a two-step restoration strategy, which produces inaccurate transmission maps and poor scene radiance for noisy and hazy inputs. We propose…

Abstract: k-space preconditioning accelerates iterative Magnetic Resonance Imaging (MRI) reconstructions from non-uniformly sampled k-space data. Reconstruction accuracy is sacrificed by sampling density compensations or circulant preconditioners. We fix both issues. We demonstrate that the dual…

Abstract: Smoke density estimation from a single image is a novel, ill-posed problem. We propose a wave-shaped neural network, W-Net, by stacking convolutional encoder-decoder structures. Stacking encoder-decoders directly increases network depth, expanding receptive fields for…

Abstract: Binary descriptors are popular for multi-view matching and robotic navigation. In non-planar scenes, severe scale and viewpoint changes reduce their matching performance. We suggest using compact spatio-temporal descriptors to encode the varying appearance of…