compression_Sky-and-Birds.py

import math from PIL import Image import numpy as np from quantizer import Quantizer from scipy import fftpack class PI: def __init__(self, path): self.image = Image.open(path) self.pixels = np.array(self.image) self.newimage = [] for i in self.pixels: pixel = [] for j in i: pixel.append(float(j)) self.newimage.append(pixel) self.newimage = np.array(self.newimage) def shownp(self, im): im = Image.fromarray(np.uint8(im)) im.show(im) def save(self, path, im): im = Image.fromarray(np.uint8(im)) im.save(path) def snr(self, org, compressed): p1 = np.power(org, 2) s1 = np.sum(p1) mi = org - compressed p2 = np.power(mi, 2) s2 = np.sum(p2) snr = 20 * math.log(pow(s1, 0.5)/(pow(s2, 0.5)), 10) return snr def entropy(self, array): dic = {} s = 0 for i in array: for j in i: if j not in dic: dic[j] = 1 else: dic[j] += 1 s += 1 entropy = 0 for i in dic: p = dic[i] / s entropy -= p * math.log(p, 2) return entropy def block_oriented_DCT(self, array, d): ll = len(array) l2 = len(array[0]) b = int(ll / d) b2 = int(l2 / d) bline = np.split(array, b) topset = [] subset = [] submap = [] for l in range(len(bline)): blocks = np.split(bline[l], b2, axis=1) for block in blocks: dct = fftpack.dct(fftpack.dct(block.T, norm='ortho').T, norm='ortho') topset.append(dct[0][0]) for i in reversed(range(1, 1)): for j in range(i + 1): subset.append(dct[j][i - j]) submap.append((j, i - j)) dt, ds = self.goingon(subset, topset) topset = self.uniqua(topset, dt) subset = self.uniqua(subset, ds) print('quantized') return self.reshape(subset, topset, submap, b, b2, d) def goingon(self, subset, topset): dt = self.getdecisions(16, topset) ds = self.getdecisions(8, subset) return dt, ds def getdecisions(self, d, a): m = (np.min(a), np.max(a) + 0.001) sec = (m[1] - m[0]) / d return [m[0] + i * sec for i in range(d + 1)] def reshape(self, subset, topset, submap, l, ll, d): subset = subset[::-1] topset = topset[::-1] submap = submap[::-1] reshaped = None for i in range(l): linematrix = None for j in range(ll): ans = np.zeros((d, d)) ans[0][0] = topset.pop() for time in range(0): value = subset.pop() index = submap.pop() ans[index[0]][index[1]] = value ans = fftpack.idct(fftpack.idct(ans.T, norm='ortho').T, norm='ortho') if linematrix is not None: linematrix = np.concatenate((linematrix, ans), axis=1) else: linematrix = np.array(ans) if reshaped is not None: reshaped = np.concatenate((reshaped, linematrix), axis=0) else: reshaped = np.array(linematrix) return reshaped def uniqua(self, a, d): l = len(a) new = [] for i in range(l): for j in range(len(d)): if d[j] <= a[i] < d[j + 1]: new.append((d[j] + d[j + 1]) / 2) return new if __name__ == '__main__': path = '/Users/linan/Downloads/newbird.png' p = PI(path) q = Quantizer() pic = p.block_oriented_DCT(p.newimage, 8) p.save('/Users/linan/Downloads/G3.png', pic) p.shownp(pic) print('SNR: ', p.snr(p.newimage, pic)) #topset = np.load('/Users/linan/PycharmProjects/compression/topset.npy') #subset = np.load('/Users/linan/PycharmProjects/compression/subset.npy') #dt, ds = p.goingon(subset,topset) #p.show(p.pixels) #d = q.getdecisions(6, p.pixels) """ array = p.newimage.flatten() print(type(array)) r, c = q.clustering_based(array, 6) print(r) print(len(c))""" #newimage2 = np.reshape(q.cde(r, c), (448, 640)) #print('E:', p.entropy(newimage2)) #print('SNR: ', p.snr(p.newimage, newimage2)) #p.save('/Users/linan/Downloads/cluster_based.png', newimage) #r1, d = q.iterate(p.newimage, d, 10) #print(r1) #for i in range(len(d)-1): # print(d[i], d[i+1], (d[i+1] + d[i]) / 2) #image = q.deq(p.newimage, d, r1) #print('E:', p.entropy(image)) #print('SNR:', p.snr(newimage2, image)) #p.save('/Users/linan/Downloads/Max_Lloyd_quantizer.png', image)