Architectural Design of an Encoder for a Visually Motivated Real-time Image Compression Scheme

Fast Image Compression by Quadtree Prediction

“…the boundaries of large blocks are too visible. Filtering has sometimes been used to fade out the disturbing edges [6,8,10,18], but other techniques have also been suggested [19]. Here, an extra filter is added to the decoding process, so that the block boundaries are smoothed out by suitable interpolation.…”

“…Our choice is variable-scale quantization, which is a widely accepted principle [6,8,9]: for small blocks, lower precision is used than for large blocks. This satisfied both visual and computational quality criteria: for high-variance regions the precise pixel values are not important, whereas for low-variance regions they are.…”

“…Warter [8], and R. Wilson [9]. We now compare the steps of our compression and decompression algorithms mainly with theirs.…”

“…The tree is built bottom-up, starting from singular pixels, because it is easy and fast to calculate the three values, based on their values in the four sub-blocks. The bottom-up approach was recommended also in [6,8], whereas topdown construction was applied in [7,9]. The sum is calculated, instead of the average, because repeated calculation of averages would create accumulating rounding errors.…”

“…A typical way is to set a threshold for the variance of pixels within the block. More sophisticated conditions have been developed [6,8]. We shall use a simpler and faster condition, namely a threshold for the difference between maximum and minimum pixel values, which has been rather commonly used, see e.g.…”