Automatic and accurate shadow detection from (potentially) a single image using near-infrared information

Shadows, due to their prevalence in natural images, are a long studied phenomenon in digital photography and computer vision. Indeed, their presence can be a hindrance for a number of algorithms; accurate detection (and sometimes subsequent removal) of shadows in images is thus of paramount importance. In this paper, we present a method to detect shadows in a fast and accurate manner. To do so, we employ the inherent sensitivity of digital camera sensors to the near-infrared (NIR) part of the spectrum. We start by observing that commonly encountered light sources have very distinct spectra in the NIR, and propose that ratios of the colour channels (red, green and blue) to the NIR image gives valuable information about impinging illumination. In addition, we assume that shadows are contained in the darker parts of an image for both visible and NIR. This latter assumption is corroborated by the fact that a number of colorants are transparent to the NIR, thus making parts of the image that are dark in both the visible and NIR prime shadow candidates. These hypotheses allow for fast, accurate shadow detection in real, complex, scenes, including soft and occlusion shadows. We demonstrate that the process is reliable enough to be performed in-camera on still mosaicked images by simulating a modified colour filter array (CFA) that can simultaneously capture NIR and visible images. Finally, we show that our binary shadow maps can be the input of a matting algorithm to improve their precision in a fully automatic manner.

Joint Acquisition of Color and Near-Infrared Images on a Single Sensor

Zahra Sadeghipoor Kermani

In the past few years, fusing NIR and color images has been explored in general computational photography and computer vision tasks, where traditionally only color images are used. The additional information provided by the differences of light and scene reflections in the visible and NIR bands of the electromagnetic spectrum is used in several applications such as image denoising, image dehazing, shadow detection and removal, and high dynamic-range imaging. In this thesis, we study a system that simultaneously captures color and NIR images on a single silicon sensor. Such a camera could be manufactured with minor changes in the hardware of consumer color cameras and it could be integrated inside small devices such as cell phones. We address two main challenges in color and NIR acquisition. First, we study the spatial and spectral sampling of the scene, which is inevitable in single-sensor acquisition of multiple spectral channels. We then focus on chromatic aberration distortions. Similarly to color imaging, we use a color filter array (CFA) to sample the scene in visible and NIR bands. We address two main challenges regarding the CFA: (1) designing the CFA, and (2) developing a demosaicing algorithm to reconstruct full-resolution images from the subsampled measurements. We consider a general CFA with filters that transmit different mixtures of color and NIR channels. We develop a framework that, by exploiting the spatial and spectral correlations of color and NIR images, computes the transmittance of each filter and the demosaicing matrix. Our optimized CFA and demosaicing outperform other solutions developed for single-sensor color and NIR acquisition. We also investigate a CFA that is formed by one blue, one green, one red, and one NIR-pass filter. We call it the RGBN CFA and assume that, similarly to color cameras, it uses dye filters that do not have sharp cut-offs. Hence, color and NIR radiations leak into NIR and color filters, respectively. We devise an algorithm that reconstructs full-resolution images from mixed and subsampled sensor measurements. The RGBN CFA and our reconstruction algorithm perform as well as or even better than other single-sensor acquisition techniques that use more complicated hardware components. The problem of chromatic aberration is caused by deficiencies of optical elements. A simple lens converges light rays with different wavelengths at different distances from the lens. Hence, if the color image is in focus and sharp on the sensor plane, the NIR image captured with the same focus settings is out of focus and blurred. We propose an algorithm that retrieves the lost details in NIR using the gradients of the sharp color image. As the high-frequency details of color and NIR images are not strongly correlated in all image patches, our method locally adapts the contribution of color gradients in deblurring. To achieve this, we develop a multiscale scheme that iterates between deblurring NIR and estimating the correlation between color and NIR high-frequency components. Our algorithm outperforms both blind and guided deblurring approaches. We also design a method that estimates a dense blur-kernel map when the severity of chromatic aberration changes as the depths of objects vary across the image. Our method performs better than the competing methods both in estimating the blur-kernel map and in deblurring.

EPFL2015

Illuminant estimation and detection using near infrared

Sabine Süsstrunk

Digital camera sensors are sensitive to wavelengths ranging from the ultraviolet (200-400nm) to the near-infrared (700-100nm) bands. This range is, however, reduced because the aim of photographic cameras is to capture and reproduce the visible spectrum (400-700nm) only. Ultraviolet radiation is filtered out by the optical elements of the camera, while a specifically designed "hot-mirror" is placed in front of the sensor to prevent near-infrared contamination of the visible image. We propose that near-infrared data can actually prove remarkably useful in colour constancy, to estimate the incident illumination as well as providing to detect the location of different illuminants in a multiply lit scene. Looking at common illuminants spectral power distribution show that very strong differences exist between the near-infrared and visible bands, e.g., incandescent illumination peaks in the near-infrared while fluorescent sources are mostly confined to the visible band. We show that illuminants can be estimated by simply looking at the ratios of two images: a standard RGB image and a near-infrared only image. As the differences between illuminants are amplified in the near-infrared, this estimation proves to be more reliable than using only the visible band. Furthermore, in most multiple illumination situations, one of the light will be predominantly near-infrared emitting (e.g., flash, incandescent) while the other will be mostly visible emitting (e.g., fluorescent, skylight). Using near-infrared and RGB image ratios allow us to accurately pinpoint the location of diverse illuminant and recover a lighting map.

2009

Enhancing the Visible with the Invisible: Exploiting Near-Infrared to Advance Computational Photography and Computer Vision

Sabine Süsstrunk

Silicon-based digital camera sensors exhibit significant sensitivity beyond the visible spectrum (400-700nm). They are able to capture wavelengths up to 1100 nm, i.e., they are sensitive to near-infrared (NIR) radiation. This additional information is conventionally treated as noise and is absorbed by a NIR-blocking filter affixed to the sensor. We show that retaining instead of removing NIR information can significantly improve certain computational photography and computer vision tasks. Indeed, intrinsic properties of the NIR wavelength band guarantee that images can be sharper, less affected by man-made colorants, and more resilient to changing light conditions. The benefits of using NIR images in conjunction with standard color images in applications such as haze removal, skin smoothing, single and multiple illuminant detection, shadow detection, and material classification is discussed.

SID2010

Joint Acquisition of Color and Near-Infrared Images on a Single Sensor

Zahra Sadeghipoor Kermani

EPFL2015