Lossy compression is all about removing information that is not vital to the image or sound. If you have an 8 × 8 array of pixels (i.e., the normal image) it is impossible to easily remove information to compress the image. If an image is transformed into DCT coefficients, the data becomes ordered in the sense that the 64 DCT coefficients are arranged from 1 to 63 with 0 being the D.C. coefficient (the average luminance) and the other coefficients corresponding to horizontal and vertical waves of increasing frequency. The highest numbered coefficients are the highest visual frequencies and these often tend to be low values (if you photograph a sky with little detail, there are no high?frequency visual components). Consequently, you can sometimes reduce these coefficients to zero with

little effect on the image.

What will be an ideal response?

Multimedia applications tend to have relatively small wordlengths; typically, 8?, 12?, or 16?bits. Moreover,
processing often involves identical operations being applied to a block of data. Suppose you have an image with one million pixels and you are enhancing its contrast by taking the actual range of 8?bit pixels, which may range
from intensities of 40 to 200, and scaling them to the full range of 0 to 255. Clearly, you can scale all one million
pixels in parallel. In practice, SIMD operation lets you take, say, eight 8?bit pixels in a register and apply eight
operations on them in parallel.