Software pixel interpolation for digital still cameras suitable for a 32-bit MCU

Abstract

This paper discusses the interpolation technique applied to the Bayer primary color method, used frequently as the pixel structure of CCD image sensors for digital still cameras. Eight typical types of interpolation methods are discussed from three viewpoints: the characteristics of the interpolated images, the processing time required to realize their methods based on a 32-bit MCU for embedded applications, and the quality of the resultant images. In terms of reducing the occurrences of pseudocolor and achieving good color restoration, the linear interpolation method taking G's correlation determined by using R/B pixels into consideration was found to be excellent. The measured machine cycle of the interpolation methods was approximately 46 cycles per pixel. Therefore, every method was able to interpolate a VGA-size image in approximately 0.2 seconds with the MCU operating at 60 MHz. In terms of the S/N ratio, a good image quality was obtained through the linear interpolation methods, even with shorter processing time. Based on these results it is concluded that the linear interpolation method, which takes correlation into consideration, is the most suitable for consumer product applications such as digital still cameras.