mpeg7_color_layout.py

import cv2
import numpy as np

# ------------------------- MPEG7 Color Layout Descriptor -------------

def divide_blocks_and_get_dominant_color(image):
    # Get Image dimensions
    height, width, channels = image.shape

    blocks = []
    # Creating an 8x8 matrix
    h = 8
    v = 8

    for i in range(v):
        for j in range(h):
            temp = image[int((height / v) * i): int((height / v) * (i + 1)),
                         int((width / h) * j): int((width / h) * (j + 1))]
            blocks.append(np.array(average_color(temp)))

    return get_larger_image(blocks, h, v, 1)


def average_color(temp):
    # Get the average color from a block
    avg_row = np.average(temp, axis=0)
    avg_color = np.average(avg_row, axis=0)
    return_img = np.ones((1, 1, 3), dtype=np.uint8)
    return_img[:, :] = avg_color
    return return_img


def get_larger_image(blkList, horizontal, vertical, times):
    overAllList = []
    for i in range(horizontal):
        eachRow = []
        for j in range(vertical):
            for k in range(times):
                eachRow.append(blkList[i * horizontal + j][0][0])
            temp = np.array(eachRow)
        for k in range(times):
            overAllList.append(temp)

    return np.array(overAllList)


def get_mpeg7_features(image):
    # Image partitioning and representative color selection
    representative_img = divide_blocks_and_get_dominant_color(image)

    # Conversion of color space from RGB to YCbCr
    YCbCr = cv2.cvtColor(representative_img, cv2.COLOR_BGR2YCrCb)

    # Get the Y, Cb and Cr components
    Y, Cb, Cr = YCbCr[:, :, 0] / 255.0, YCbCr[:, :, 1] / \
        255.0, YCbCr[:, :, 2] / 255.0

    # DCT transformation of each component
    dctY = cv2.dct(Y)
    dctCb = cv2.dct(Cb)
    dctCr = cv2.dct(Cr)

    # Zigzag scanning of the transformed matrices\
    scannedDctY = np.concatenate(
        [np.diagonal(dctY[::-1, :], i)[::(2 * (i % 2) - 1)] for i in range(1 - dctY.shape[0], dctY.shape[0])])
    scannedDctCb = np.concatenate(
        [np.diagonal(dctCb[::-1, :], i)[::(2 * (i % 2) - 1)] for i in range(1 - dctCb.shape[0], dctCb.shape[0])])
    scannedDctCr = np.concatenate(
        [np.diagonal(dctCr[::-1, :], i)[::(2 * (i % 2) - 1)] for i in range(1 - dctCr.shape[0], dctCr.shape[0])])

    return (scannedDctY, scannedDctCb, scannedDctCr)


def execute_mpeg7(dataset):
    features = []

    for img in dataset:
        features.append(get_mpeg7_features(img.img))

    feature_arr = np.array(features)

    return feature_arr