A method was developed to restore degraded images to some extent after the pixel binning process in image sensors to improve the resolution. A pixel binning model was used to approximate the original un-binned image. Then, the least squares error criterion was used as a constraint to reconstruct the restored pixel values from the binning model. The technique achieves about a one-decibel increase in the peak signal-to-noise ratio compared with the original estimated image. The technique has good detail preservation performance as well as low computation load. Thus, this restoration technique provides valuable improvements in practical, real time image processing.

This paper presents an optimized implementation of the FDK algorithm on a single fixed-point TMS320C6455 digital signal processor (DSP). Software pipelining and proper configuration of the data transfer enables a 2563 volume to be reconstructed in about 42 seconds from 360 projections with very good accuracy. This implementation reveals the potential of modern high-performance DSPs in accelerating image reconstruction, especially when cost and power consumption are emphasized.