0

DCT of 1x8 data [8,16,24,32,40,48,56,64] --(dct8)--> [100,-52,0,-5,0,-2,0,0.4]. Now if we truncate (i.e. discard low-frequency components) and then take Inverse DCT of that 1x4 data over 1x8 interval as : [100,-52,0,-5] --(idct8)--> [8,15,24,32,40,48,57,63] . i.e. we almost get the same data.

Can this be used to reduce size of jpeg images by applying 2D DCT, followed by discarding low frequency values (because most of signal's energy lies @high frequency)i.e. discarding the bottom-right quadrant, followed by 2D IDCT?

P.S. : I have to work on 'text recognition' so a lossy compression won't affect my performance rather it could be much much faster.

Nitin Kotwar
  • 1
  • 3

1 Answers1

1

Unlike most compression scheme, JPEG uses a series of compression step. What you are describing is one of them that JPEG already uses.

user3344003
  • 20,574
  • 3
  • 26
  • 62
  • I know JPEG does this kind processing but then again if I apply this same algorithm to any already compressed jpeg then there is a decrement in size, sometime upto 80%, again information will be lost, but I don't need those noisy pixels anyway for detecting text regions, so **will it work** ? – Nitin Kotwar Oct 30 '16 at 06:29
  • Doing it a second time will not improve compression. – user3344003 Oct 31 '16 at 04:12