联系方式

  • QQ:99515681
  • 邮箱:99515681@qq.com
  • 工作时间:8:00-23:00
  • 微信:codinghelp

您当前位置:首页 >> Java编程Java编程

日期:2020-04-21 10:28

CS4551 Multimedia Software Systems

Homework2 () – Vector Quantization and DCT Coding


?What to turn in:

oSubmit source code with necessary files for “compile and run”.

oDo NOT submit data files.

oYou MUST provide a readme.txt file containing all information to help with the grading process.

?If your program produces any compile errors, you will receive 0 automatically no matter how close your program is to the solution.


?Programming requirements:

oYou are not allowed to use any Java built-in image class methods, library, or tools to complete this homework.

oDo not create one mega-size main class.

oDo not change any given methods of MImage class nor create a new class that duplicates MImage

class. Treat MImage as a part of imported library.

oTest your program with all test data.

oIf you do not meet any of the requirements above, you will receive a significant reduction.

0.What your program should do

Name your main application CS4551_[YourLastName].java (e.g. CS4551_Doe.java) and expand the given template program to perform the following tasks.

Receive the input file as command line arguments.

<eg> On Command Prompt

java CS4551_Doe Ducky.ppm

Read a 24bit input PPM image and display the following main menu to the user and receive the user’s input.


Main  Menu

1.Vector  Quantization

2.DCT-based  Coding

3.Quit


Please  enter  the  task  number  [1-3]:


After performing a selected task, go back to display the menu again.


1.Task 1 – Vector Quantization (50 pts)

Compress the 24 bits per pixel input image to 2 bits per pixel using Vector Quantization (VQ). Implement VQ encoding/decoding using the requirements below.


Encoding:

?Input vectors are formed by 2×2 blocks of RGB pixels. Each input vector ?? consists of RGB values of FOUR pixels, P1, P2, P3, and P4, and therefore ?? is 12 dimensional.


P1P2

P3P4

Diagram of a 2×2 pixel block of the input image

?? = {??1??, ??1??, ??1??, ??2??, ??2??, ??2??, ??3??, ??3??, ??3??, ??4??, ??4??, ??4??}

?Codebook and codebook vectors: The 2-bits per pixel quantization is equivalent to using 8 bits per 4 pixels. Therefore, the VQ should the vector space into 256 (=28) cells and the codebooks should have 256 entries that are centroids of the 256 cells. After the vector quantization, each vector ?? belongs to one cell and each cell number is represented by 1 byte. In order words, after the quantization, each 2×2 block (4×3=12 bytes) is encoded by a 1-byte codebook index. So, the compression ratio is 12.

?Codebook generation: Use K-means clustering algorithm to generate codebook vectors (centroids of cells).

K-means Clustering Algorithm

Inputs: K, number of clusters and the data set (input vectors ??) K is 256 in our case.

Assume that ??[??] store the K centroids. ?? = 0, 1, ? , 255. Each ??[??] is a 12-dimensional vector.

1.Assign randomly generated initial values for the ?? centroids.

2.For each ??,

For each ?? = 0 to 255

If ??[??] is  the  closest  cell  (cluster)  to ?? based  on  the  Euclidean  distance  between ?? and ??[??], assign ?? to ??[??] cluster

3.Update the ?? centroids.

4.Iterate 2 & 3 until the algorithm meets a stopping condition (i.e. no data points change clusters, the sum of the distance is minimized, or the maximum number of iterations is reached.)

?Display the codebook. This is equivalent to displaying ??[??],  ?? = 0, 1, ? ,255.

?Extra credit (10 pts) – Save the quantized image (i.e. indices of all 2×2 blocks) into a PPM file. Given

?? × ?? input image, the quantized image resolution is ??/2 × ??/2. The quantized image is a grayscale image because each pixel is an 8-bit index.

Decoding:

?Given the quantized image and the codebook, for each pixel of the quantized image, recover RGB pixel values of 4 pixels.

?Save the decoded image so that you can compare the output with the input.


2.Task 2 – DCT-based Coding (50 pts)

Implement a DCT-based transform coding. Notice that this is different from the standard JPEG steps. The encoder/decoder steps are shown below.



Input RGB Image (PPM)Decompressed RGB Image (PPM)

Encoding StepsDecoding Steps

E1. Read and resize the input image

Read the input ppm file containing RGB pixel values for encoding. First, if the image size is not a multiple of 8 in each dimension, make (increase) it become a multiple of 8 and pad with zeros. For example, if your input image size is 21×14, make it become 24×16 and fill the extra pixels with zeros (black pixels).D4. Remove Padding and Display the image

Display the decompressed image. Remember that you padded with zeros if the input image size is not multiple of 8 in both dimensions (width and height). Restore the original input image size by removing extra padded rows and columns.


E2. Color space transformation and Subsampling

Transform each pixel from RGB to YCbCr using the equation below:

??0.29900.58700.1140??

(????) = (?0.1687?0.33130.5000 ) (??)

????0.5000?0.4187?0.0813??

Initially, RGB value ranges from 0 and 255. After color transformation, Y should range from 0 to 255, while Cb and Cr should range from -127.5 to 127.5. (Truncate if necessary.)

Subtract 128 from Y and 0.5 from Cb and Cr so that they span the same range of values [-128,127]

Subsample Cb and Cr using 4:2:0 (MPEG1) chrominance subsampling scheme. If Cb(Cr) is not divisible by 8, pad with zeros.D3. Supersampling and Inverse Color space transformation

Supersample Cb and Cr so that each pixel has Cb and Cr.

Add 128 to the values of the Y component and 0.5 to the values of the Cb and Cr components.

If using a color image, transform from the YCbCr space to the RGB space according to the following equation:

??1.000001.4020??

(??) = (1.0000?0.3441?0.7141) (????)

??1.00001.77200????

Common mistake: After this step, you have to make sure that the resulting RGB values are in the range between 0 and 255. Truncate if necessary.

E3. Forward DCT

Perform the forward DCT for Y image using the following steps:

?Divide the image into 8×8 blocks. Scan each block in the image in raster order (left to right, top to bottom)

?For each 8×8 block, perform the DCT transform to get the values ?????? from the values

??????. The elements ?????? range from ?210 to 210

Check max and min and assign ?210 or 210 for the values outside of the range so that the values range from ?210 to 210.

Perform the DCT for Cb and Cr images, too.D2. Inverse DCT

Perform the inverse DCT to recover the values ?????? from the values ?????? and recover Y, Cb, Cr images.


Forward DCT Formula

1,otherwise

?????? is the ??-th row and ??-th column pixel of the 8×8 image block (?? and ?? range from 0 to 7); ?????? is the DCT coefficient value in the ??-th row and ??-th column (?? and ?? range from 0 to 7).

Inverse DCT Formula

E4. Quantization

Given ?????? in an 8×8 DCT block, quantize ?????? using:

Quantized(?? ) = round ( ?????? )

??????????

The default intervals ?????? corresponding ?? and ?? are specified in Table 1 and Table 2.

In this homework, we want to provide a variety of compression quality options (high compression or low compression). User shall specify one parameter

?? ( 0 ≤ ?? ≤ 5 ), which controls the quality of the compression by changing the quantization intervals. The actual quantization is done by

?????? Quantized(??????) = round (??′   )

????

??′  = ??× 2??

????????

For example, if ?? = 0, ??′ is same as ?? ; if ?? = 1,

????????

??′ is double of ?? , which will divide ?? with

????????????

bigger values and result in more compression.D1. De-quantization

Assume that the quantization tables (basis ones) and the compression quality parameter ?? are available for decoding. Given the quantized value for DCT coefficient ?????? , multiply it by the corresponding quantization interval ??′ .

????

??′ = Quantized(?? ) × ??′

????????????

Notice that the recovered ??′ will be different from

????

the original ??????.

?E1/D4 – 10 pts

?E2/D3 – 10 pts

?E3/D2 – 20 pts

?E4/D1 – 10 pts

An important requirement – After each encoding step, implement the corresponding decoding step immediately and check if your output is correct or not.

You will receive credits for each encoding step if only if you complete to implement the corresponding decoding step.


Sample results will be posted on CSNS.


版权所有:留学生编程辅导网 2020 All Rights Reserved 联系方式:QQ:99515681 微信:codinghelp 电子信箱:99515681@qq.com
免责声明:本站部分内容从网络整理而来,只供参考!如有版权问题可联系本站删除。 站长地图

python代写
微信客服:codinghelp