Data adaptive filters for demosaicking: a framework

Abstract

A new demosaicking framework for single-sensor imaging devices operating on a Bayer color filter array (CFA) is introduced and analyzed. An efficient data adaptive filtering concept in conjunction with the refined spectral models constitutes the base for the proposed framework. Using a different form of the function mapping the aggregated absolute differences among the CFA inputs to the edge-sensing weighting coefficients, the framework allows to design fully automated demosaicking solutions suitable for common digital imaging apparatus, and alternatively, the proposed solutions can also be used to support PC-based demosaicking of the raw CFA images. Thus, the framework can be seen as a universal tool satisfying the needs of the end-users for i) the instant access and visualization of the captured images, and ii) the interactive processing of the raw sensor data. Moreover, the proposed framework is relatively easy to implement in either software or hardware. Experimental results indicate that the proposed framework exhibits excellent performance in terms of the commonly used objective criteria and at the same time it produces demosaicked images with impressive visual quality.