def index_image(id, img):
im = Image.open(img)
arr = np.asarray(im, dtype='float')
row, col = arr.shape[0], arr.shape[1]
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321], [0.212, -0.523, 0.311]])
arr = np.dot(arr, tranform.T)
images = [arr[:,:,:1].reshape(row, col), arr[:,:,1:2].reshape(row, col), arr[:,:,2:].reshape(row, col)]
avgl = []
haars = []
colors = 3
num_coefs = 40
imgbuckets = dbspace["buckets"]
for i in range(colors):
haar = haar2d(images[i]).reshape(row*col)
avgl.append(haar[0]/(256*128))
haar[0] = 0.0
ind = np.argpartition(np.absolute(haar), 0-num_coefs)[-num_coefs:]
haar = haar[ind]
small = (haar <= 0)
big = (haar > 0)
origin_small = ind[small]
origin_big = ind[big]
for j in origin_small:
imgbuckets[i][0][j].append(id)
for j in origin_big:
imgbuckets[i][1][j].append(id)
dbspace["sigs"][id] = avgl
def test_haar2d2(img):
im = Image.open(img)
arr = np.asarray(im, dtype='float')
row, col = arr.shape[0], arr.shape[1]
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321], [0.212, -0.523, 0.311]])
arr = np.dot(arr, tranform.T)
images = [arr[:,:,:1].reshape(row, col), arr[:,:,1:2].reshape(row, col), arr[:,:,2:].reshape(row, col)]
colors = 3
haars = [haar2d(images[i]).reshape(row*col) for i in range(colors)]
avgl = [0.0, 0.0, 0.0]
avgl = [haars[i][0]/(256*128) for i in range(colors)]
for i in range(colors):
haars[i][0] = 0.0
print 'avgl', avgl
lefts = 40
inds = [np.argpartition(np.absolute(haars[i]), 0-lefts)[-lefts:] for i in range(colors)] #lefts
haars = [haars[i][inds[i]] for i in range(colors)] #value in lefts
big_i = [(haars[i] > 0) for i in range(colors)]
small_i = [(haars[i] < 0) for i in range(colors)]
for i in range(colors):
print inds[i][big_i[i]]
print inds[i][small_i[i]]
def test_haar2d():
"""
Asserts the forwards and inverse wavelet transformations are correct.
"""
assert haar2d(np.random.random([5,3]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape."
assert haar2d(np.random.random([8,4]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape, if required."
image = np.random.random([5,3])
haart = haar2d(image, 3)
haari = ihaar2d(haart, 3)[:image.shape[0], :image.shape[1]]
assert (image - haari < 0.001).all(), "Transform must be circular."
def test_haar2d():
"""
Asserts the forwards and inverse wavelet transformations are correct.
"""
assert haar2d(np.random.random([5,3]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape."
assert haar2d(np.random.random([8,4]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape, if required."
image = np.random.random([5, 3])
haart = haar2d(image, 3)
haari = ihaar2d(haart, 3)[:image.shape[0], :image.shape[1]]
assert (image - haari < 0.001).all(), "Transform must be circular."
def query_img(img):
imgbuckets = dbspace["buckets"]
sigs = dbspace["sigs"]
#TODO do in a function
im2 = Image.open(img)
im = im2.resize((128, 128))
arr = np.asarray(im, dtype='float')
row, col = arr.shape[0], arr.shape[1]
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321],
[0.212, -0.523, 0.311]])
arr = np.dot(arr, tranform.T)
images = [
arr[:, :, :1].reshape(row, col), arr[:, :, 1:2].reshape(row, col),
arr[:, :, 2:].reshape(row, col)
]
avgl = []
haars = []
colors = 3
num_coefs = 40
img_sig = [[0] * 2 for i in range(colors)]
for i in range(colors):
haar = haar2d(images[i]).reshape(row * col)
avgl.append(haar[0] / (256 * 128))
haar[0] = 0.0
ind = np.argpartition(np.absolute(haar), 0 - num_coefs)[-num_coefs:]
haar = haar[ind]
small = (haar <= 0)
big = (haar > 0)
img_sig[i][0] = ind[small]
img_sig[i][1] = ind[big]
scores = {}
for id, sig in sigs.items():
scores[id] = 0.0
for c in range(colors):
scores[id] = scores[id] + (weights[0][c] *
math.fabs(sig[c] - avgl[c]))
for pn in range(2):
for c in range(colors):
for idx in img_sig[c][pn]:
for uit in imgbuckets[c][pn][idx]:
scores[uit] = scores[uit] - weights[img_bin[idx]][c]
scores = sorted(scores.items(), key=lambda item: -item[1])
results = []
for k, v in scores:
results.append(k)
results.append(v)
return normalizeResults(results)
def query_img(img):
imgbuckets = dbspace["buckets"]
sigs = dbspace["sigs"]
#TODO do in a function
im2 = Image.open(img)
im = im2.resize((128,128))
arr = np.asarray(im, dtype='float')
row, col = arr.shape[0], arr.shape[1]
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321], [0.212, -0.523, 0.311]])
arr = np.dot(arr, tranform.T)
images = [arr[:,:,:1].reshape(row, col), arr[:,:,1:2].reshape(row, col), arr[:,:,2:].reshape(row, col)]
avgl = []
haars = []
colors = 3
num_coefs = 40
img_sig = [[0]*2 for i in range(colors)]
for i in range(colors):
haar = haar2d(images[i]).reshape(row*col)
avgl.append(haar[0]/(256*128))
haar[0] = 0.0
ind = np.argpartition(np.absolute(haar), 0-num_coefs)[-num_coefs:]
haar = haar[ind]
small = (haar <= 0)
big = (haar > 0)
img_sig[i][0] = ind[small]
img_sig[i][1] = ind[big]
scores = {}
for id,sig in sigs.items():
scores[id] = 0.0
for c in range(colors):
scores[id] = scores[id] + (weights[0][c] * math.fabs(sig[c] - avgl[c]))
for pn in range(2):
for c in range(colors):
for idx in img_sig[c][pn]:
for uit in imgbuckets[c][pn][idx]:
scores[uit] = scores[uit] - weights[img_bin[idx]][c]
scores = sorted(scores.items(), key = lambda item: -item[1])
results = []
for k,v in scores:
results.append(k)
results.append(v)
return normalizeResults(results)
def index_image(id, img_file):
img = Image.open(img_file)
size = (128, 128)
im = img.resize(size)
arr = np.asarray(im, dtype='float')
row, col = arr.shape[0], arr.shape[1]
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321],
[0.212, -0.523, 0.311]])
arr = np.dot(arr, tranform.T)
images = [
arr[:, :, :1].reshape(row, col), arr[:, :, 1:2].reshape(row, col),
arr[:, :, 2:].reshape(row, col)
]
avgl = []
haars = []
colors = 3
num_coefs = 40
imgbuckets = dbspace["buckets"]
for i in range(colors):
haar = haar2d(images[i]).reshape(row * col)
avgl.append(haar[0] / (256 * 128))
haar[0] = 0.0
ind = np.argpartition(np.absolute(haar), 0 - num_coefs)[-num_coefs:]
haar = haar[ind]
small = (haar <= 0)
big = (haar > 0)
origin_small = ind[small]
origin_big = ind[big]
#print "i is :", i, "neg is:"
#print origin_small
#print "pos is :"
#print origin_big
for j in origin_small:
imgbuckets[i][0][j].append(id)
for j in origin_big:
imgbuckets[i][1][j].append(id)
print avgl
dbspace["sigs"][id] = avgl
assert haar2d(np.random.random([5,3]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape."
assert haar2d(np.random.random([8,4]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape, if required."
image = np.random.random([5,3])
haart = haar2d(image, 3)
haari = ihaar2d(haart, 3)[:image.shape[0], :image.shape[1]]
assert (image - haari < 0.001).all(), "Transform must be circular."
if __name__ == '__main__':
#test_haar2d()
#http://pda.readthedocs.org/en/latest/chp4.html
# , duli de weidu
image = np.random.random([8,8]) * 10
#image = np.asarray([83.000000, 53.000000, 93.000000, 63.000000, 82.000000, 71.000000, 44.000000, 49.000000, 51.000000, 33.000000, 63.000000, 12.000000, 78.000000, 98.000000, 34.000000, 39.000000], dtype=float)
print image
haar = haar2d(image)
#print haari
ihaar = ihaar2d(haar)
print ihaar
assert (image - ihaar < 0.001).all(), "Transform must be circular."
assert haar2d(np.random.random([5,3]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape."
assert haar2d(np.random.random([8,4]),2,debug=True).shape == (8,4), \
"Transform data must be padded to compatible shape, if required."
image = np.random.random([5, 3])
haart = haar2d(image, 3)
haari = ihaar2d(haart, 3)[:image.shape[0], :image.shape[1]]
assert (image - haari < 0.001).all(), "Transform must be circular."
if __name__ == '__main__':
#test_haar2d()
#http://pda.readthedocs.org/en/latest/chp4.html
# , duli de weidu
image = np.random.random([8, 8]) * 10
#image = np.asarray([83.000000, 53.000000, 93.000000, 63.000000, 82.000000, 71.000000, 44.000000, 49.000000, 51.000000, 33.000000, 63.000000, 12.000000, 78.000000, 98.000000, 34.000000, 39.000000], dtype=float)
print image
haar = haar2d(image)
#print haari
ihaar = ihaar2d(haar)
print ihaar
assert (image - ihaar < 0.001).all(), "Transform must be circular."
def test_haar2d(img):
im = Image.open(img)
#im.show()
arr = np.asarray(im, dtype='float')
#plt.imshow(arr, cmap = cm.Greys_r)
#plt.show()
arr = arr/255
#arr = arr[0:5,0:5]
arr2 = arr.copy()
row, col = arr.shape[0], arr.shape[1]
assert (arr - arr2 < 0.0001).all()
tranform = np.array([[0.299, 0.587, 0.114], [0.596, -0.275, -0.321], [0.212, -0.523, 0.311]])
#print arr[0,0]
#print np.dot(arr[0,0], tranform.T)
#print colorsys.rgb_to_yiq(*arr[0,0])
arr = np.dot(arr, tranform.T)
arr_r,arr_g,arr_b = (np.zeros([row, col]), np.zeros([row, col]), np.zeros([row, col]))
arr3 = arr.copy()
for i in range(row):
for j in range(col):
r,g,b = colorsys.rgb_to_yiq(*arr2[i,j])
arr_r[i,j] = r
arr_g[i,j] = g
arr_b[i,j] = b
arr3[i,j] = [r,g,b]
assert (arr - arr3 < 0.01).all()
images = np.array([arr[:,:,:1].reshape(row, col), arr[:,:,1:2].reshape(row, col), arr[:,:,2:].reshape(row, col)])
assert (images[0] - arr_r < 0.01).all()
colors = images.shape[0]
haars = [haar2d(images[i]) for i in range(colors)]
#print haars[0].shape
#print row, col
#print haars[0] - images[0]
assert not (images[0] - haars[0] < 0.1).all()
haars = [haars[i].reshape(row*col) for i in range(colors)]
lefts = 41
inds = [np.argpartition(np.absolute(haars[i]), 0-lefts)[:((row**2)-lefts)] for i in range(colors)]
print inds[0].shape
#reversed_inds = [list(set(range(row**2)) - set(inds[i])) for i in range(colors)]
for i in range(colors):
haars[i][inds[i]] = np.zeros(inds[i].shape[0])
haars = [haars[i].reshape([row, col]) for i in range(colors)]
ihaars = [ihaar2d(haars[i]) for i in range(colors)]
#assert (images[0] - ihaars[0] < 0.1).all()
for i in range(row):
for j in range(col):
r,g,b = colorsys.yiq_to_rgb(ihaars[0][i,j], ihaars[1][i,j], ihaars[2][i,j])
arr3[i,j] = [r,g,b]
arr3 = arr3*255
#arr3 = arr3.astype(numpy.int32, copy=False)
#plt.imshow(arr3, cmap = cm.Greys_r)
#plt.show()
img = Image.new('RGB', [row,col])
pixels = img.load()
for i in range(row):
for j in range(col):
pixels[j,i] = (int(arr3[i,j][0]), int(arr3[i,j][1]), int(arr3[i,j][2]))
img.show()