Face detection

gistfile1.py

# Use local scikit-image
import sys
sys.path.insert(0, "/home/dan/University/projects/gsoc_face_detection/scikit-image/")

from skimage.feature import multiblock_local_binary_pattern
from skimage.transform import integral_image
import numpy as np
import skimage.io as io
import xml.etree.ElementTree as ET


EPS = 1e-5

tree = ET.parse('lbpcascade_frontalface.xml')

# Load all the features
features = tree.find('.//features')

# Load stages
stages = tree.find('.//stages')

stages_amount = int(tree.find('.//stageNum').text)

#img_big = io.imread('face_big_1.jpg', as_grey=True)


def detect_face(img):

    integral_img = integral_image(img)

    for stage_number in xrange(stages_amount):

        current_stage = stages[stage_number]

        stage_threshold = float(current_stage.find('stageThreshold').text)

        weak_classifiers_amount = int(current_stage.find('maxWeakCount').text)
        weak_classifiers = current_stage.find('weakClassifiers')

        stage_points = 0

        for weak_classifier_number in xrange(weak_classifiers_amount):

            current_weak_classifier = weak_classifiers[weak_classifier_number]

            # Weights that will be added to current image score.
            # If image is probably to have face the positive weight is
            # added. Otherwise, negative.
            leaf_values = current_weak_classifier.find('leafValues').text
            leaf_values = map(lambda x: float(x), leaf_values.split())


            # Extract the elements only starting from second.
            # First two are useless
            internal_nodes = current_weak_classifier.find('internalNodes')
            internal_nodes = internal_nodes.text.split()[2:]

            # Extract the feature number and respective parameters.
            # The MBLBP position and size.
            feature_number = int(internal_nodes[0])
            feature_params = features[feature_number][0].text.split()
            feature_params = map(lambda x: int(x), feature_params)

            # Compute the MB-LBP with extracted params
            lbp_code = multiblock_local_binary_pattern(integral_img, *feature_params)
            #skimage.feature.visualize_multiblock_lbp(img, *feature_params, **{'lbp_code': lbp_code})

            # Extract look-up table
            lut = map(lambda x: int(x), internal_nodes[1:])
            lut = np.asarray(lut, dtype='uint32')

            # Get the bit that shows whether or not
            # current image contains a face.
            bit = (lut[lbp_code >> 5] >> (lbp_code & 31)) & 1

            # Add respective leaf value
            stage_points += leaf_values[0] if bit else leaf_values[1]

        if stage_points < (stage_threshold - EPS):

            return False

    return True