Home | People | Research | Publications | Software | Teaching


Computational Photography


Super Resolution and High Dynamic Range Imaging

Super-resolution restoration combines multiple images to improve spatial resolution. By intentionally changing the exposure parameters during the video capture, we can introduce photometric diversity and obtain images with both higher dynamic range and higher spatial resolution. [Read more]


High Dynamic Range Image Registration

High dynamic range image requires registration of differently exposed images. This is a challenging problem because most optical flow estimation algorithms depend on constant brightness assumption. In this paper, we estimate dense motion field and non-parametric photometric mapping from images with different exposures.  [Read more]

Coherent Imaging


Multi-Transmitter Aperture Synthesis

Multi-transmitter aperture synthesis increases the effective aperture in coherent imaging by shifting the backscattered speckle field across a physical aperture or set of apertures. Through proper arrangement of the transmitter locations, it is possible to obtain speckle fields with overlapping regions, which allows fast computation of optical aberrations from wavefront differences. In this paper, we present a method where Zernike polynomials are used to model the aberrations and high-order aberrations are estimated without the need to phase unwrapping of the difference fronts. [Read more]


Sparse Aperture Coherent Imaging

The resolution of a diffraction-limited imaging system is inversely proportional to the aperture size. Instead of using a single large aperture, small multiple apertures are used to synthesize a large aperture. Such a multi-aperture system requires phasing sub-apertures to within a fraction of a wavelength in order to work. In this paper, we present an approach to correct for piston, tip, tilt, as well as rotational and translational errors. [Read more]

Inverse Problems


Super-Resolution Imaging

Super-resolution image restoration is the process of producing a high-resolution image (or a sequence of images) from a set of low-resolution images. The process requires an image acquisition model that relates a high-resolution image to multiple low-resolution images and involves solving the resulting inverse problem. We have conducted research addressing several issues, including modeling, registration, compression, and color sampling. [Project page]



Noise characteristics in an image depend on many factors, including sensor type, pixel dimensions, temperature, exposure time, and ISO speed. Most image denoising algorithms are optimized for white Gaussian noise and may not work as well with real data. We have developed a denoising framework that combines the strengths of multi-resolution denoising approach and the bilateral filter. [Read more]



Popular image and video coding standards are based on block processing of image data. This results in block discontinuities in addition to other compression artifacts. These artifacts could be reduced by using an in-loop or post-processing deblocking filter. We developed an adaptive deblocking filter that can effectively reduce artifacts while preserving edges and texture. [Read more]



Demosaicking is a key step in a digital camera pipeline. We developed a demosaicking algorithm that effectively uses intra- and inter-channel correlations in an alternating optimization scheme. [Read more]

Computer Vision Tools


Fast Bilateral Filter

The bilateral filter is used in a wide range of applications in image processing and computer vision. One drawback of the bilateral filter is the high computational complexity. In this project, we present a fast bilateral filter implementation with good speed and accuracy. The method uses multiple box  kernels and optimally combines them to approximate an arbitrary domain kernel. [Read more]


Illumination Robust Feature  Extraction

Interest point detection is a necessary task in many computer vision applications. A good interest point detection algorithm should be robust to geometric transformations, noise, and illumination changes. We developed an algorithm that can improve the illumination robustness of most interest point detectors. We theoretically show that the algorithm improves the illumination robustness of Harris corner detector; in our experiments, the repeatability rate of the Harris corner detector under large illumination changes is improved by around 25%. [Read more]