matlab 读取耶鲁YaleFace的pgm文件保存为mat

YaleFace人脸数据集包含丰富的人脸数据，包含各种光照条件，在深度学习的人脸识别和光度立体三维重建中应用广泛。数据集参考链接如下：

1. Yale-Face-database
2. ExtYaleDatabase
   原文件为.info和.pgm格式，需要读取转换格式。
   
   P00: 是人脸的标号；
   A-30: A是azimuth，即方位角，所对应的数值为-30°；
   E+40: E是elevation，即仰角，所对应的数值为+40°。
   方位角、天顶角、仰角的关系如下图所示：
   仰角为天顶角的余角。通过方位角和天顶角描述方向向量参考该篇
   $$\begin{gathered} 设仰角elevation记为{\theta }_l; \\ 方位角azimuth记为{\phi }_l \end{gathered}$$
   不难得到光源方向L的表达式：
   $$\vec{L}=\left(\cos {\theta }_l\cos {\phi }_l,\cos {\theta }_l\sin {\phi }_l,\sin {\theta }_l\right)$$
   现利用matlab读取数据转存到mat格式：

clc
close all
clear all

datapath = './yaleB12/';
dstpath = './yaleB12_mat/';
type = '*.info';

if(~exist(dstpath))
    mkdir(dstpath)
end

infolists = dir([datapath type]);
infonums = length(infolists);
for info_id = 1:infonums
    infoname = infolists(info_id).name;
    infodatas = importdata([datapath infoname]);
    lens = length(infodatas);  % 直接对文件夹下.pgm进行检索也可
    imglists_480_640 = [];  % 480*640为实际图像大小，保存到imglists中会转换为列向量
    lightlists = [];
    for img_id = 1:lens
        imgname = infodatas{img_id};
        if(~startsWith(imgname,'yaleB'))
            continue
        end
        if contains(imgname,'Amb')
            imgAmb = importdata([datapath imgname]);
            imglists_480_640 = [imglists_480_640,imgAmb(:)];
            continue
        end
        imgface = importdata([datapath imgname]);
        azimuth = str2double(imgname(13:16));
        elevation = str2double(imgname(18:20));
        lightdirection = Genlightvector(elevation,azimuth);
        imglists_480_640 = [imglists_480_640,imgface(:)];
        lightlists = [lightlists;lightdirection];
    end
    save([dstpath strrep(infoname,'info','mat')],"imglists_480_640","lightlists")
end

function lightvector = Genlightvector(elevation,azimuth)
    lightvector = [cosd(azimuth) * cosd(elevation) sind(azimuth) * cosd(elevation) sind(elevation)];
end

读取保存的mat文件，可以查看图像信息：

clc
close all
clear all

datapath = './yaleB11_mat/';
type = '*.mat';
matlists = dir([datapath type]);
matnums = length(matlists);
for i = 1:matnums
    imglists_480_640 = load([datapath matlists(i).name]).imglists_480_640;
    lightlists = load([datapath matlists(i).name]).lightlists;
    imgnums = size(imglists_480_640,2);
    img_step = 4;
    nums = imgnums / img_step;
    for j = 1:nums
        figure
        for img_id = 1:img_step
            imgface = imglists_480_640(:,img_step * (j - 1) + img_id);
            imgface = reshape(imgface,480,640);
            subplot(2,2,img_id)
            imshow(imgface)
        end
        pause(1)
    end
end