jpeg_quality_score.py

#!/usr/bin/python
# -*- coding: utf-8 -*-

import sys
import numpy as np
import Image


def jpeg_quality_score(filename):
    img = np.array(Image.open(filename))

    M, N = img.shape
    if M < 16 or N < 16:
        return -2

    x = np.double(img)

    ###########################################################################
    # Feature Extraction:
    ###########################################################################

    # 1. horizontal features

    d_h = x[:, 1:N] - x[:, 0:(N - 1)]

    B_h = np.mean(abs(d_h[:, 7:(8 * (np.floor(N / 8) - 1)):8]))

    A_h = (8 * np.mean(abs(d_h)) - B_h) / 7

    sig_h = np.sign(d_h)
    left_sig = sig_h[:, 0:(N - 2)]
    right_sig = sig_h[:, 1:(N - 1)]
    Z_h = np.mean((left_sig * right_sig) < 0)

    # 2. vertical features

    d_v = x[1:M, :] - x[0:(M - 1), :]

    B_v = np.mean(abs(d_v[7:(8 * (np.floor(M / 8) - 1)):8, :]))

    A_v = (8 * np.mean(abs(d_v)) - B_v) / 7

    sig_v = np.sign(d_v)
    up_sig = sig_v[0:(M - 2), :]
    down_sig = sig_v[1:(M - 1), :]
    Z_v = np.mean((up_sig * down_sig) < 0)

    # 3. combined features

    B = (B_h + B_v) / 2
    A = (A_h + A_v) / 2
    Z = (Z_h + Z_v) / 2

    ###########################################################################
    # Quality Prediction
    ###########################################################################

    alpha = -245.8909
    beta = 261.9373
    gamma1 = -239.8886 / 10000
    gamma2 = 160.1664 / 10000
    gamma3 = 64.2859 / 10000

    score = alpha + beta * (B ** gamma1) * (A ** gamma2) * (Z ** gamma3)
    return score


if __name__ == "__main__":
    #print jpeg_quality_score("Sego_lily_cm_nb.jpg")
    #print jpeg_quality_score("Sego_lily_cm-150_nb.jpg")
    print jpeg_quality_score(sys.argv[1])