内容简介:Using DCT is for extracting the approximation of the photo and omit detail, because human eyes don't perceive changes too much in detail, so .Jpeg uses this idea to filter out the details of the photo for reducing the photo size.DCT is similar to Fourier t
JPEG-DCT-Compression
catalog │── 1. Preset │ │── 1.1 Read image, set bits occupied by "Run-Length", use RGB-to-YCbCr or not and block size. │ └── 1.2 Transfer RGB image to YCbCr Image if you want │── 2. Transform the image into blocks │── 3. DCT (Discrete Cosine Transform) │ │── 3.1 operate DCT on all the blocks respectively, check the difference between step-2 and step-3 │ │── 3.2 Why DCT (0,0) or (0) is called DC? and why it is the most light point? │ └── 3.3 How it looks like transforming the DCT blocks back to image? │── 4. Quantized DCT(save approximations, omit details) │ │── 4.1 Use default quantization table(you can find it in google) to quantize all dct blocks │ │── 4.2 Transform the quantized dct blocks back to image, check the difference between step-3.1 and step-4.2 │ │── 4.3 Try a random bad quantization table to check the difference of result image. │ └── 4.4 Decompression(image reconstruction) of 3 kinds of DCT(check difference) │── 5. Example of saving one quantized DCT block as small size as it can be │ │── 5.1 Zig-Zag process │ │── 5.2 Huffman Coding │ │── 5.3 Run-Length encoding(with VLI Variable-length-integer encoding) │ └── 5.4 Save the result of Run-Length to bytes │── 6. Saving as jpeg, do step-5 for all blocks of the image │── 7. Decoding │ │── 7.1 Run-Length decoding │ └── 7.2 Zig-Zag decoding │ └── 7.2.1 Another Zig-Zag Decoding └── 8. Loading the jpeg, decompression of compressed(encoded) image
Basic idea of DCT
Using DCT is for extracting the approximation of the photo and omit detail, because human eyes don't perceive changes too much in detail, so .Jpeg uses this idea to filter out the details of the photo for reducing the photo size.
DCT is similar to Fourier transform, below is doing Fourier transform on the whole picture and then generate its high-pass and low-pass pictures.
Run all Code cells from beginning to end
These are the parameters you could modify:
w=8 #modify it if you want, maximal 8 due to default quantization table is 8*8 h=w runBits=1 #modify it if you want bitBits=3 #modify it if you want useYCbCr=True #modify it if you want useHuffman=True #modify it if you want quantizationRatio=1 #quantization table=default quantization table * quantizationRatio
For above default parameters, the result is below
I only focus on explanations, so didn't improve duration that much:
Image original size: 2.250 MB
Compression image size: 0.110 MB
Compression ratio: 20.41 : 1
Encoding: 12.198627233505249 seconds
Decoding: 2.6317562580108643 seconds
Original image: Compression image:
以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网
猜你喜欢:- 图像处理 20:1 压缩,逐步推导 jpeg 格式压缩过程,包括数学原理解析
- 四元数公式推导
- 线性回归数学推导
- Lucene打分公式的推导
- 现代 C++:自动类型推导
- 机器学习分享——反向传播算法推导
本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们。
用户体验草图设计
比尔·巴克斯顿(Bill Buxton) / 黄峰 / 电子工业出版社 / 2009-11 / 168.00元
《用户体验草图设计:正确地设计,设计得正确(全彩)》:比尔·盖茨亲笔推荐版 人因国际、百度、华为、微软、腾讯用户体验部门联合推荐!一起来看看 《用户体验草图设计》 这本书的介绍吧!