Browse Prior Art Database

Gamut Constrained Illumination Estimation

IP.com Disclosure Number: IPCOM000022186D
Publication Date: 2004-Mar-01
Document File: 5 page(s) / 86K

Publishing Venue

The IP.com Prior Art Database

Related People

Graham D. Finlayson: INVENTOR [+3]

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 41% of the total text.

         Sony Corporation

         Sony Electronics Inc.

         IPD Case #50R4768

Title:

Gamut Constrained Illumination Estimation

Inventors:

Graham D. Finlayson

Steven Hordley

Ingeborg Tastl

Description of the Invention:

1.      An accurate estimation of the illumination of a scene is essential for the performance of an Automatic White Balancing Algorithm of a Digital Still Camera. Failures results in colorcasts of the final images and can clearly been perceived by consumers.

The colors recorded by a digital still camera depend both on the surface reflectance properties of the objects of the scene and on the spectral power distribution of the incident light. However, since the human visual system. to a large degree compensates for the color of the illumination, we see colors that are approximately constant whatever the illumination. This implies that when developing a color reproduction pipeline for a digital still camera it is necessary to correct for the color of the light illuminating the scene. Good correction is possible provided that the scene illumination is known, but in an application as digital photography it is very often not known and must therefore be estimated from the image data. Many approaches to this estimation problem have previously been reported in the literature [12,3,14,4,16,19,9,5,10,6,2,17] each with its own strengths and weaknesses. A failing shared by many of the existing approaches is the fact that they do not work reliably on real images and are thus unsuitable for use in a commercial camera. This failure is mostly due to making unrealistic assumptions about scene composition (e.g. every scene contains a white surface). But weaker assumptions lead to more complex algorithms [9,10]. Not only are these algorithms more difficult to implement but they have not been proven to be robust in all imaging conditions. Indeed some of the weakest assumptions lead to very poor performance [16,18,14]. An exception to this is the color by correlation algorithm by Finlayson, Hordley and Hubel [6, 11] which provides good performance for most images. But this algorithm, which is patented. is clearly not the end of the line for illumination estimation research.

This invention builds on the best features of some of the more successful existing algorithms [9,6,17] and is in particular been inspired by the work of Forsyth [9].

In it's original form Forsyth's algorithm can often give a good estimate of the scene illumination and on average it performs better than previous approaches such as Grey-World [3] or Max-RGB[12]. The algorithm does though suffer from a number of limitations. First, the success of the algorithm depends on the degree to which illumination change can be modeled by a diagonal matrix transform. Second, even when the properties of the device sensors imply that such a model is in general a good one, real images often contain features (for example specularities) which confound the model. In either of the two cases, the algorithm will often return no...