Color filter array with sparse color sampling crosses for mobile phone image sensors

In order to solve the conflict between the size constraints of mobile phone cameras and image quality, it is an effective approach to improve the light sensitivity of imaging sensors. A color filter array comprising sparsely distributed crosses of color sampling blocks is proposed, in which nearly 69% of elements are transparent ones that detect luminance directly with a higher sensitivity than the remaining color-filtered ones. Monochromatic images resulted from the majority white pixels are of high spatial resolution and of low noise. They are used along with the sparse color samplings to generate color images. An advantage of the color sampling crosses is that monochromatic images can be estimated with less error, which is critical for the successive calculation of color images. Experiments have been conducted using a Canon 30D camera under normal and low light levels. Results showed that the color filter array could greatly reduce image noise that usually occurs with the conventional Bayer pattern under low light levels, and did not cause noticeable color artifacts that would normally occur with undersampling of chrominance. Introduction Phone cameras are much more readily to use than standalone digital cameras. As mobile phones are becoming smaller and thinner, the size constraints of mobile phones put forward a request of small image sensors. At the same time, consumers are continuously pursuing higher image quality. It is a well-known fact that the reduction in sensor size causes a reduction in light sensitivity, and therefore may cause a reduction in image quality if compensating illumination is not provided, for example, by flashlights. However, when phone cameras are used as video cameras, using flashlight usually is not a possible option. In addition, market surveys show that mobile phone cameras are frequently needed under low light levels. In order to solve the conflict between sensor size and image quality, to improve the light sensitivity of sensors is an effective approach. Comparing with a black-and-white camera, a color camera using the same image sensor and a RGB color filter array (CFA) has only 1/10 of the ISO speed of the black-and-white one [1]. Some CyanMagenta-Yellow CFAs and RGB CFAs with white pixels have been proposed in an attempt to increase light energy reaching image sensors [2, 3]. However, the CMY CFAs usually have color artifact problems, and those CFAs with white pixels usually provide limited improvements because of low rates of white pixels (<50%). A fundamental problem in these CFAs is that luminance and chrominance is not really separated. Hence, improvements in one generally cause losses in another. A novel imaging paradigm with majority pixels being white ones and the remaining pixels being sparsely distributed color filtered ones has been proposed by the author [4], in which white pixels generate luminance information and RGB color pixels provide pure chrominance information. Luminance and chrominance are integrated in Lab color space. Because luminance and chrominance can be separated well in sampling and processing stages, the rate of white pixels can be designed to be over 50%. In the novel paradigm, luminance is very important because it determines the resolution of output color images. Luminance can also provide useful information for chrominance calculation as there is a strong correlation between luminance and chrominance for natural images. Therefore, it is critical to make a good estimation of luminance. A CFA comprising sparse color sampling crosses is proposed in this paper, with which good estimation of luminance can be likely.