Showing papers by "Hang Zhao published in 2015"

Loss Functions for Neural Networks for Image Processing

Abstract: Neural networks are becoming central in several areas of computer vision and image processing and different architectures have been proposed to solve specific problems. The impact of the loss layer of neural networks, however, has not received much attention in the context of image processing: the default and virtually only choice is L2. In this paper, we bring attention to alternative choices for image restoration. In particular, we show the importance of perceptually-motivated losses when the resulting image is to be evaluated by a human observer. We compare the performance of several losses, and propose a novel, differentiable error function. We show that the quality of the results improves significantly with better loss functions, even when the network architecture is left unchanged.

Unbounded High Dynamic Range Photography Using a Modulo Camera

Abstract: This paper presents a novel framework to extend the dynamic range of images called Unbounded High Dynamic Range (UHDR) photography with a modulo camera. A modulo camera could theoretically take unbounded radiance levels by keeping only the least significant bits. We show that with limited bit depth, very high radiance levels can be recovered from a single modulus image with our newly proposed unwrapping algorithm for natural images. We can also obtain an HDR image with details equally well preserved for all radiance levels by merging the least number of modulus images. Synthetic experiment and experiment with a real modulo camera show the effectiveness of the proposed approach.

Is L2 a Good Loss Function for Neural Networks for Image Processing

Abstract: Neural networks are becoming central in several areas of computer vision and image processing. Different architectures have been proposed to solve specific problems. The impact of the loss layer of neural networks, however, has not received much attention by the research community: the default and most common choice is L2. This can be particularly limiting in the context of image processing, since L2 correlates poorly with perceived image quality. In this paper we bring attention to alternative choices. We study the performance of several losses, including perceptually-motivated losses, and propose a novel, differentiable error function. We show that the quality of the results improves significantly with better loss functions, even for the same network architecture.

Methods and apparatus for virtual sensor array

Abstract: A time-of-flight camera images an object around a corner or through a diffuser. In the case of imaging around a corner, light from a hidden target object reflects off a diffuse surface and travels to the camera. Points on the diffuse surface function as a virtual sensors. In the case of imaging through a diffuser, light from the target object is transmitted through a diffusive media and travels to the camera. Points on a surface of the diffuse media that is visible to the camera function as virtual sensors. In both cases, a computer represents phase and intensity measurements taken by the camera as a system of linear equations and solves a linear inverse problem to (i) recover an image of the target object; or (ii) to compute a 3D position for each point in a set of points on an exterior surface of the target object.

Snapshot On-chip HDR ROIC Architectures

Abstract: We describe novel digital readout integrated circuits (DROICs) that achieve snapshot on-chip high dynamic range imaging where most commercial systems require a multiple exposure acquisition.