Scribd
Upload
67 views4 pages

EEL707 - Multimedia Systems: Assignment 2

This document analyzes the effect of different block sizes (2x2, 4x4, 8x8, 16x16) when applying the discrete cosine transform (DCT) to compress an image. The reconstructed images using smaller block sizes are more blurred with larger compression artifacts compared to larger block sizes. The quality increases with larger block sizes up to 16x16 due to rejecting fewer high frequency components. However, beyond a size of 32x32, the root mean square error increases again due to losing delicate image details when encoding with very large blocks. In summary, the appropriate block size depends on the image characteristics, with smaller blocks better preserving details but introducing more artifacts.

Uploaded by

Shashank Kushwah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
67 views4 pages

EEL707 - Multimedia Systems: Assignment 2

This document analyzes the effect of different block sizes (2x2, 4x4, 8x8, 16x16) when applying the discrete cosine transform (DCT) to compress an image. The reconstructed images using smaller block sizes are more blurred with larger compression artifacts compared to larger block sizes. The quality increases with larger block sizes up to 16x16 due to rejecting fewer high frequency components. However, beyond a size of 32x32, the root mean square error increases again due to losing delicate image details when encoding with very large blocks. In summary, the appropriate block size depends on the image characteristics, with smaller blocks better preserving details but introducing more artifacts.

Uploaded by

Shashank Kushwah
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

EEL707 Multimedia Systems

Assignment 2
-Vinod Kumar (2012EE50566)

Question:Apply DCT transform to an Image for different block sizes taking four different cases: 2x2,
4x4 (smaller); 8x8(Standard); 16x16 (Larger).
Also comment on the variation in the quality of output images.

Method:

The input image used is of resolution 256x256 pixels.


The inbuilt Discrete Cosine Transform (DCT) in MATLAB is used and applied to
image.
The different block size considered are:
2x2, 4x4, 8x8, 16x16
The mask rejects the high frequency components and allow low frequency
components around DC value.
Only 25% of the DCT coefficients, which are centred around DC value get passed
through the mask.
The output of the mask is used to reconstruct the image by applying Inverse Discrete
Cosine Transform.

Output Images:1. Reference (Input Image):

2. Output Image with block Size 2x2:-

3. Output Image with block Size 4x4:-

4. Output Image with block Size 8x8:-

5. Output Image with block Size 16x16:-

Discussion:Comparing the output images with input image it is observed that the output images are
relatively blurred and do not have sharp edges. This is because the Discrete Cosine transform
of image is passed through a low pass filter and hence the high frequency components are
rejected. Thus on reconstructing the images via Inverse DCT, some details of the image are
lost which causes cause blurring of an image. Hence it can be verified that the DCT is a lossy
compression because of the presence of compression artifacts in the reconstructed images
which are generated due to rejection of high frequency components by masks.
On comparing the output for different block sizes, it can be concluded that the quality of image
increases with increase in block size. The output image with block size 2x2 if of minimum
quality while the image with block size 16x16 is of maximum quality.
The reasons for this variation in quality of image with block size is due to the difference in the
blocking artifacts and different RMS Error. The images reconstructed out of smaller block size
have large number of compression artifacts present in them while image reconstructed out of
larger block size i.e., 8x8 and 16x16 have relatively smaller compression artifacts. The other
reason for variation in quality is RMS Error. Since for smaller blocks large number of high
frequency coefficients are rejected, so the root mean square error decreases as the size of block
is increased. But this is not so beyond the block size higher than 32x32. It is observed that if
we increase the block size beyond 32x32, there RMS Error also increases.
For a smaller block size, there are large number of pixels which takes up same value in
reconstruction and frequency components with finer variations are rejected. This leads to a
large value of RMS Error in reconstructed image. While for the large block, the frequency
component with low content variation are not rejected, hence the reconstructed image contains
more details and hence less blur. But beyond the block size 32x32, the RMS Error increases
because of the fact that on encoding high frequency element with very large block size, the
delicate image details are lost.
Hence this discussion can concluded by saying that as per the requirement different block size
can be used. In case where there is low correlation between the image pixels the smaller block
size can be used so that the high frequency significant components are not lost. While if the
image pixel vales are highly correlated, then large block can be used for compression.

You might also like