Digital Image Processing project-new Color Balancing Method for Cameras
The problem of separating the illumination from the re ectance information in a given image has been extensively researched in the last three decades, following Edwin Land’s seminal work on color vision and his development of the Retinex theory [4]. The problem can be described as follows given an input image S, we would like to decompose it into two dierent images the re ectance image R and the illumination image L, such that S (x; y) = R (x; y) L (x; y). There are many benefits to such a decomposition, including the ability to correct for color-shifts due to illumination, correct for uneven illumination, introduce artificial lighting and enhancing dynamic range. It is not hard to see that in general, this problem is ill-posed for a given input image L, there are infinitely possible solutions of L and R pairs that can explain S. Many works have tried to constraint the problem, by posing assumptions on the type of illumination (e.g. constant-hue illumination over the field-of-view and spatial smoothness). With the growing popularity of digital cameras the importance of fast algorithms for color correction (also known as auto white-balancing, AWB in short) grew as well. Such algorithms are an integral part of the image signal processing (ISP) pipeline that is responsible for converting the RAW image captured by the sensor into the final color JPEG image that is saved on the memory card. AWB algorithms try to estimate the correct three whitebalance gains (for the red, green and blue channels) that should be applied on an input image in order to correct for color shifts caused by illumination, so that white elements in the scene indeed appear white in the image similar to the way the human visual system can compensate for dierent lighting conditions so that white color always seems white under dierent illuminations. Figure 1 shows an example of correct vs. incorrect white balancing.
Free download research paper
CSE PROJECTS