These three tables show the effects of increasing quantization. In each table, the figures in the green boxes show the original values as might be given after a DCT operation. The figures in yellow are the coefficients applied.
The figures in pink boxes show how the data is reduced after dividing the coefficients into the original values, and rounding the results. Finally the blue boxes show the results when the picture is decompressed, that is the pink figures multiplied by the yellow figures.


Original Values 235 127 74 22 8 3
Quantization Coefficient 14 13 12 11 10 9
Quantized Values 17 10 6 2 1 0
Decompressed Values 238 130 72 22 10 0

The table above shows the effect of lowish quantization values, 14, 13, 12 etc. The decompressed figures show a maximum error of 3, i.e. a pretty accurate result.


Original Values 235 127 74 22 8 3
Quantization Coefficient 32 30 28 26 24 22
Quantized Values 7 4 3 1 0 0
Decompressed Values 224 120 84 26 0 0

The table above shows the effect of higher quantization values, 32, 30, 28 etc. The decompressed figures show a maximum error of 11. This is about 5% of the most significant values, and the least significant values have been reduced to zero.


Original Values 235 127 74 22 8 3
Quantization Coefficient 60 56 52 48 44 40
Quantized Values 4 2 1 0 0 0
Decompressed Values 240 112 52 0 0 0

The table above shows the effect of even higher quantization values, 60, 56, 52 etc. Note that the most significant value is still quite accurate, in fact more accurate than the previous result. This demonstrates that choosing the quantization coefficients is a black art! More of the less significant values have been eliminated, in fact the figures in green have been reduced to just 4, 2 and 1, so vast amounts of data have been lost. However the reconstituted picture is still not too far removed from the original.