MIPI RAW图像数据与RAW图像数据格式介绍与读写代码

在ISP图像处理中经常需要面对MIPI RAW数据和RAW数据，这里简单说明一下二者的差别。
传感器采集的RAW数据通常为10bit，存储RAW数据需要两个Byte，而其中有6个bit位是空着的，这样就有存储空间浪费。MIPI RAW数据充分利用了这个特性，采用5个Byte，共40bit存储4个RAW数据。这里以大端存储方式，给出这样存储的示意图。如下图所示：

从MIPI RAW数据恢复RAW数据的方法：

//b1,b2,b3,b4,b5分别为连续读出的五个uint8类型的数据，p1,p2,p3,p4为四个像素点的像素值，那么
p1 = (b1 << 2) + ((b5     ) & 0x3);
P2 = (b2 << 2) + ((b5 >> 2) & 0x3);
P3 = (b3 << 2) + ((b5 >> 4) & 0x3);
P4 = (b4 << 2) + ((b5 >> 6) & 0x3);
1
2
3
4
5

这里给出raw mipi10的读写matlab 代码示例，方便读者理解和认识mipi raw图像数据与unpacked raw数据

function rawImg = imreadRaw(fileName, width, height, isMipi)
% read raw 10/12/14/16 or raw mipi 10 from a binary file

fid = fopen(fileName, 'r');

if isMipi == 1
    imgSize = width * height * 5 / 4;
    imgData = fread(fid, imgSize, 'uint8');
    rawImg = zeros(width*height, 1);
    
    % byte0 | byte1 | byte2 | byte3 | byte4
    % byte 0/1/2/3 store high 8 bits of pixel value, and byte4 store2
    % low 2 bits of each pixel value
    
    % byte 0
    rawImg(1:4:end) = imgData(1:5:end);
    % byte 1
    rawImg(2:4:end) = imgData(2:5:end);
    % byte 2
    rawImg(3:4:end) = imgData(3:5:end);
    % byte 3
    rawImg(4:4:end) = imgData(4:5:end);
    % byte 4
    rawImgLow2Bits  = imgData(5:5:end);
    
    rawImg = rawImg * 4;
    
    rawImg(1:4:end) = rawImg(1:4:end) + bitand(bitshift(rawImgLow2Bits, 0), 3);
    rawImg(2:4:end) = rawImg(2:4:end) + bitand(bitshift(rawImgLow2Bits, -2), 3);
    rawImg(3:4:end) = rawImg(3:4:end) + bitand(bitshift(rawImgLow2Bits, -4), 3);
    rawImg(4:4:end) = rawImg(4:4:end) + bitand(bitshift(rawImgLow2Bits, -6), 3);
else
    rawImg = fread(fid, width * height, 'uint16');
end

fclose(fid);

rawImg = reshape(rawImg, width, height)';
    
figure, imshow(uint16(rawImg), [0, 1023])

end
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42

function imwriteRaw(fileName, rawImg, isMipi)

[height, width] = size(rawImg);
rawImg = rawImg';
rawImg = rawImg(:);
fid = fopen(fileName, 'w');

rawImg = uint16(rawImg);

if isMipi == 1
    rawData = zeros(height * width * 5 / 4);
    % pix0 
    pix0 = rawImg(1:4:end);
	pix0Low2Bits = bitand(pix0, 3);
    % pix1
    pix1 = rawImg(2:4:end);
	pix1Low2Bits = bitand(pix1, 3);
    % pix2
    pix2 = rawImg(3:4:end);
	pix2Low2Bits = bitand(pix2, 3);
    % pix3
    pix3 = rawImg(4:4:end);
	pix3Low2Bits = bitand(pix3, 3);
    
    byte0 = uint8(pix0 / 4);
    byte1 = uint8(pix1 / 4);
    byte2 = uint8(pix2 / 4);
    byte3 = uint8(pix3 / 4);
	
	% big endian mipi store
    byte4 = bitshift(pix3Low2Bits, 6) + bitshift(pix2Low2Bits, 4) + ...
			bitshift(pix1Low2Bits, 2) + pix0Low2Bits;
    
    rawData(1:5:end) = byte0;
    rawData(2:5:end) = byte1;
    rawData(3:5:end) = byte2;
    rawData(4:5:end) = byte3;
    rawData(5:5:end) = byte4;
    
    fwrite(fid, uint8(rawData), 'uint8');
else
    fwrite(fid, uint16(rawImg), 'uint16');
end

fclose(fid);

end
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47

如下链接打包了读写raw数据及mipi raw数据的代码及仿真数据，方便读者下载复现和学习raw图像数据的存储格式和读写过程，如果出现问题，可以给笔者发私信沟通
https://download.csdn.net/download/luzhanbo207/33239267