"""
Extract HOG feature of images.But we decide to use C instead of Python considering the 10-times efficiency gap.

@author: chunk
chunkplus@gmail.com
2014 Dec
"""
__author__ = 'hadoop'

from mfeat import *

from common import *
import numpy as np
from PIL import Image
import cv2
from skimage.feature import hog
from skimage import io, color, transform, exposure


class FeatHOG(FeatureBase):
    def __init__(self, size=(64, 64)):
        FeatureBase.__init__(self)
        self.size = size

    def _feat_sk(self, image, size=(64, 64)):
        """
        N.B. image[x][y][z] where x for line and y for column,z(0,1,2) for (R,G,B)
        """
        if size is None:
            size = (64, 64)
        img = io.imread(image, as_grey=True)  # 0.02s
        img = transform.resize(img, size)
        # img = color.rgb2gray(img)

        fd = hog(img, orientations=9, pixels_per_cell=(8, 8),
                 cells_per_block=(2, 2), visualise=False, normalise=False)  # 0.18s
        # hog_image_rescaled = exposure.rescale_intensity(hog_image, in_range=(0, 0.02))
        #
        # cv2.imshow('processed', ski2cv(hog_image_rescaled))
        # cv2.waitKey(0)

        return fd


    def _feat_cv(self, image, size=(64, 64)):
        """
        N.B. image[x][y][z] where x for line and y for column,z(0,1,2) for (B,G,R)

        HOGDescriptor(_winSize, _blockSize, _blockStride, _cellSize, _nbins[, _derivAperture[, _winSigma[, _histogramNormType[, _L2HysThreshold[, _gammaCorrection[, _nlevels]]]]]])
        """
        if size is None:
            size = (64, 64)
        img = cv2.imread(image, cv2.CV_LOAD_IMAGE_GRAYSCALE)  # 0.0s
        img = cv2.resize(img, size, interpolation=cv2.INTER_LINEAR)

        params = dict(
            _winSize=size,
            _blockSize=(16, 16),
            _blockStride=(8, 8),
            _cellSize=(8, 8),
            _nbins=9,
            _derivAperture=1,
            _winSigma=-1,
            _histogramNormType=cv2.HOGDESCRIPTOR_L2HYS,
            _L2HysThreshold=0.2,
            _gammaCorrection=False,
            _nlevels=cv2.HOGDESCRIPTOR_DEFAULT_NLEVELS,
        )
        # feat_size = 1764 = (64/8-1)^2 * 2^2 * 9

        hog = cv2.HOGDescriptor(**params)
        desc = hog.compute(img)  # 0.03s

        return desc[:, 0]

    def set_size(self, size):
        self.size = size

    def feat(self, image):
        return self._feat_cv(image, self.size)