The original image used in the compression as well as the DCT transform of the image are displayed below.
The masked version of the DCT is below, mask 1 was used to highlight the difference from the unmasked version. As you can see, the picture has been reduced to an array of single points, each 7 pixels apart.
The results of this compression algorithm using masks 1, 2, 3 and 4 are shown below in that order. Charts comparing similarity and mask number, and compression and similarity are also included after the images.
As one can see from the images, the visual quality drops off fairly quickly below mask number 4, but the similarity never drops below 85% percent, even when only 1 of every 64 bits is retained. Additionally, the compression ratio can never be above 8, as this is the compression ratio achieved by using the all 0 mask given that 0’s require 1 bit to represent while other numbers with more digits require 8. Ultimately however, this method is clearly able to achieve very good compression without high visual loss, up to 7.2X smaller with at least 85% of the information retained.
Image Compression 