Transform Domain
Chi‐Wah Kok, Wing‐Shan Tam · 2018
Transform domain image processing can be applied in image interpolation for both algorithmic efficiency and quality improvement. This chapter considers a simpler problem of interpolating a 1D sampled signal in the Fourier domain to visualize how transform domain processing can be applied in image interpolation. The 2D discrete cosine transform (DCT) interpolation theorem suggests the application of zero padding in DCT domain to perform image interpolation. When using zero-padded block-based transformation to interpolate images, the main problem is the blocking artifacts. The blocking artifact is observed to behave like random white noise in nature but is translation variant. The chapter discusses discuss extendible inverse discrete cosine transform (EIDCT), which performs interpolated inverse discrete Fourier transform (IDFT) over DCT transformed signal. When both blocked zero-padded DCT interpolated image and EIDCT-based blocked zero-padded DFT interpolated image are available, a better image can be obtained by computing the average image from these two images.