Facial Features Detection

A program to detect facial feature points using Haarcascade classifiers for face, eyes, nose and mouth.

The sample demonstrates facial feature points detection using Haarcascade classifiers. The program detects a face and eyes, nose and mouth inside the face. The code has been tested on the Japanese Female Facial Expression (JAFFE) database and found to give reasonably accurate results.

The classifiers for face and eyes can be downloaded from:

The classifiers for nose and mouth can be downloaded from:

Sources:

function facial_features_demo(im)
    % load input image of a face
    if nargin < 1
        im = fullfile(tempdir(), 'face.jpg');
        if exist(im, 'file') ~= 2
            if true
                url = 'https://upload.wikimedia.org/wikipedia/commons/c/c9/Julia_Roberts_Cannes_2016_3.jpg';
            elseif true
                url = 'https://upload.wikimedia.org/wikipedia/commons/3/33/Nicolas_Cage_2011_CC.jpg';
            else
                url = 'https://upload.wikimedia.org/wikipedia/commons/8/8d/George_Clooney_2016.jpg';
            end
            urlwrite(url, im);
        end
        img = cv.imread(im, 'Color',true);
    elseif ischar(im)
        img = cv.imread(im, 'Color',true);
    else
        img = im;
    end

    % haarcascade classifier files for face, eye, nose, and mouth detection
    % (only the face is required, others are optional)
    xml_face = fullfile(mexopencv.root(),'test','haarcascade_frontalface_alt.xml');
    xml_eye = fullfile(mexopencv.root(),'test','haarcascade_eye_tree_eyeglasses.xml');
    xml_nose = fullfile(mexopencv.root(),'test','haarcascade_mcs_nose.xml');
    xml_mouth = fullfile(mexopencv.root(),'test','haarcascade_mcs_mouth.xml');

    % Detect faces and facial features
    out = detectFacialFeatures(img, xml_face, xml_eye, xml_nose, xml_mouth);
    imshow(out), title('Facial Features')
end

function out = detectFacialFeatures(img, xml_face, xml_eye, xml_nose, xml_mouth)
    % input image as grayscale
    out = img;
    if size(img,3) == 3
        img = cv.cvtColor(img, 'RGB2GRAY');
    end

    % detect faces
    opts = {'ScaleFactor',1.15, 'MinNeighbors',3, ...
        'ScaleImage',true, 'MinSize',[30 30]};
    faces = cascadeDetect(img, xml_face, opts{:});

    % for each face, detect facial features
    opts = {'ScaleFactor',1.20, 'MinNeighbors',5, ...
        'ScaleImage',true, 'MinSize',[30 30]};
    for i=1:size(faces,1)
        % draw face
        face = faces(i,:);
        out = cv.rectangle(out, face, 'Color',[0 0 255], 'Thickness',2);

        % eyes, nose and mouth will be detected inside the face (ROI)
        roi = cv.Rect.crop(img, face);

        % minimum object center height, used to filter detections
        % (mouth is expected to lie below nose, which in turn is below eyes)
        ymin = 0;

        % detect and draw eyes
        if ~isempty(xml_eye)
            eyes = cascadeDetect(roi, xml_eye, opts{:});
            [out, ymin] = drawDetections(out, face, eyes, ymin, [0 255 0]);
        end

        % detect and draw nose
        if ~isempty(xml_nose)
            nose = cascadeDetect(roi, xml_nose, opts{:});
            [out, ymin] = drawDetections(out, face, nose, ymin, [0 255 255]);
        end

        % detect and draw mouth
        if ~isempty(xml_mouth)
            mouth = cascadeDetect(roi, xml_mouth, opts{:});
            [out, ymin] = drawDetections(out, face, mouth, ymin, [255 0 255]);
        end
    end
end

function rects = cascadeDetect(img, xml_file, varargin)
    % if missing, attempt to download XML file from two possible locations
    if exist(xml_file, 'file') ~= 2
        [~, f, ext] = fileparts(xml_file);
        url = 'https://cdn.rawgit.com/opencv/opencv_contrib/3.2.0/modules/face/data/cascades/';
        [~,status] = urlwrite([url f ext], xml_file);
        if status == 0
            url = 'https://cdn.rawgit.com/opencv/opencv/3.2.0/data/haarcascades/';
            [~,status] = urlwrite([url f ext], xml_file);
        end
        assert(status == 1, 'Failed to download');
    end

    % detect and return Nx4 array of rectangles
    cascade = cv.CascadeClassifier(xml_file);
    rects = cascade.detect(img, varargin{:});
    rects = cat(1, rects{:});
end

function [out, ymin] = drawDetections(out, face, rects, ymin, clr)
    % filter detections
    ycenter = rects(:,2) + rects(:,4)/2;
    if ymin > 0
        idx = ycenter > ymin;
    else
        idx = true(size(rects,1),1);
    end
    ymin = max(ycenter(idx,:));

    % draw rectangle enclosing detection
    r = bsxfun(@plus, rects(idx,:), [face(1:2) 0 0]);
    out = cv.rectangle(out, r, 'Color',clr, 'Thickness',2);

    % draw center of detection
    c = bsxfun(@plus, rects(idx,1:2) + rects(idx,3:4)/2, face(1:2));
    out = cv.circle(out, c, 3, 'Color',clr, 'Thickness','Filled');
end

Warning: Image is too big to fit on screen; displaying at 67%