Hardware-efficient system for hybrid-bilinear image scaling

Abstract

A hybrid bilinear scaling (Qscale) scheme produces output images that have comparable quality to traditional bilinear interpolation algorithms, but requires a less complex hardware implementation. The Qscale system does not reverse-map output pixels back to arbitrary locations in the input space as defined by the mapping function. Rather all pixel values and locations are calculated after all of the original input pixels are mapped to the output. That is, all of the original image pixels are used “as-is” in the resultant scaled image. New pixels are generated from the original input pixels to meet the desired output pixel dimensions. Because only new pixels are computed, the Qscale system is less computationally complex. The computational requirements are further reduced because new pixels are computed between original pixel pairs meaning only two pixels are involved in the computation. Coefficients can be chosen to be fractional powers of two (0.5, 0.25, 0.125, etc) for the interpolation calculation between pixel pairs. By selecting coefficients this way, the linear computation reduces to a “shift-and-add” operation, which is easily implemented in hardware.