Image Processing Projects

Abstract: Hand-held plenoptic cameras may benefit from light field (LF) stitching. LF stitching methods cannot register scenes with large depth variation. This paper proposes a new LF stitching method to handle parallax in LFs more…

Abstract: MRI can delineate craniomaxillofacial (CMF) bony structures without radiation, unlike CT. MRI blurs bony boundaries, so structural information must be borrowed from CT during training. Paired MRI-CT data are scarce, making this difficult. We…

Abstract: Medical image registration can quantify disease progression, study longitudinal and cross-sectional data, and guide computer-assisted diagnosis and treatment. Deformable registration for retinal optical coherence tomography (OCT) images is not well developed. This paper proposes…

Abstract: Denoising smooths noisy image super-resolution (SR), making it difficult. IBP can improve the reconstructed SR image, but there is no clean reference image. This paper introduces a noisy image SR back-projection algorithm. It seeks…

Abstract: Natural image denoising priors include non-local self similarity (NSS). Most denoising methods use NSS patches. This paper introduces a pixel-level NSS prior to search similar pixels across a non-local region. Finding closely similar pixels…

Abstract: Fabric defect detection intrigues but challenges. Due to the complexity of fabric textures and defects, many fabric defect detection methods are still suboptimal. This paper proposes a GAN-based fabric defect detection framework. Considering real-world…

Abstract: Haze, smoke, fog, rain, and snow degrade outdoor images/videos. Due to weather, haze is most common in outdoor scenes. This paper proposes MSRL-DehazeNet, a deep learning-based architecture for single image haze removal using MSRL…

Abstract: CNNs have been used to study the single image super-resolution problem. Most CNN-based methods learn a model to map a downsampled low-resolution (LR) image to a high-resolution (HR) image. In general, when the down-sampling…

Abstract: Many non-local means (NLM) methods either use Euclidean distance to measure patch similarity, compute weight ij only once and keep it unchanged during denoising iterations, or use only the structure information of the denoised…

Abstract: This study applies Bayesian statistics to remote sensing imagery multimodal change detection (CD). We formulate the multimodal CD problem in unsupervised Markovian terms. The bitemporal heterogeneous satellite image pair and pixel pairwise modeling observation…

Abstract: Low-level image processing has struggled with image restoration (IR). For visually pleasing results, image priors are crucial. This paper develops a multi-channel and multi-model-based denoising autoencoder network as image prior for IR problem. First,…

Abstract: Using data ranking and linear operators, the bitonic filter distinguishes signal bitonicity (one local extremum within a set range) from noise. A circular mask locally constrained image processing. A novel structurally varying bitonic filter…