Camera functionality in modern mobile terminals : software FIR filter for demosaicing with RISC

Abstract

This thesis discusses implementation and optimizing of the image pre-processing operation demosaicing on a typical hardware architecture for mobile terminals. Work has shown that the demosaicing operation is equivalent to the mathematical operation of interpolation, and that optimizing potential lies within the implementation of the anti-imaging filter for the interpolator. This filter can be realized as an FIR filter with the filter coefficients obtained from a suitable interpolation function (kernel). Through literature review, nearest neighbor replication, bilinear and cubic convolution kernels were implemented and optimized according to a typical hardware architecture found in mobile terminals. Typical hardware architecture used consisted of ARM710T RISC processor with 8 KB on-chip cache and external SDRAM memory. The work has shown that the cubic convolution interpolation kernel offers the best image quality requiring only moderate computational effort. Though, dependent the application area, and taking the whole camera system into consideration, bilinear interpolation might be better suited as interpolation kernel for the demosaicing operation in the mobile terminal. The bilinear interpolation offers a good trade-off between the visual result and computational effort. The nearest neighbor replication interpolation kernel offers the poorest image quality, but is the most computational efficient of the kernels.