def FindTemplate(pyramid, template, threshold):
#Get dimensions of the template
tempWidth, tempHeight = template.size
#Resize template to the suggested size in order to speed up NCC computations
template = template.resize(
(suggestedTemplateWidth,
(tempHeight * suggestedTemplateWidth / tempWidth)), Image.BICUBIC)
#Call NCC and get Cross Correlation Coefficients for every image in the pyramid
crossCorrCoefficients = []
for image in pyramid:
value = ncc.normxcorr2D(image, template)
crossCorrCoefficients.append(value)
#Once you have the NCC stuff use that to find faces by checking every pixel
k = 0
#Loop over Cross Correlation Coefficients List of every image in the pyramid
for correlationArray in crossCorrCoefficients:
#access elements of pixel correlation matrix
for i in range(len(correlationArray)):
for j in range(len(correlationArray[0])):
#if correlation at the specific pixel is above threshold then draw a rectangle around the face
if correlationArray[i][j] > threshold:
draw = ImageDraw.Draw(pyramid[0])
#Draw rectangle around a detected face
draw.line(((j - tempWidth / 2) / (0.75**k),
(i - tempHeight / 2) / (0.75**k),
(j + tempWidth / 2) / (0.75**k),
(i - tempHeight / 2) / (0.75**k)),
fill="red",
width=2)
draw.line(((j - tempWidth / 2) / (0.75**k),
(i - tempHeight / 2) / (0.75**k),
(j - tempWidth / 2) / (0.75**k),
(i + tempHeight / 2) / (0.75**k)),
fill="red",
width=2)
draw.line(((j - tempWidth / 2) / (0.75**k),
(i + tempHeight / 2) / (0.75**k),
(j + tempWidth / 2) / (0.75**k),
(i + tempHeight / 2) / (0.75**k)),
fill="red",
width=2)
draw.line(((j + tempWidth / 2) / (0.75**k),
(i + tempHeight / 2) / (0.75**k),
(j + tempWidth / 2) / (0.75**k),
(i - tempHeight / 2) / (0.75**k)),
fill="red",
width=2)
#convert rectangle to color red
pyramid[0] = pyramid[0].convert('RGB')
del draw
k = k + 1
#show the original image
pyramid[0].show()
return 0
def FindTemplate(pyramid, template, threshold):
# Maintain a variable so we know by how much to scale the coordinates back.
scaleBack = 1
matches = []
# This markerImage will be the image where we draw the rectangles of matches we find.
markerImage = pyramid[0]
# Convert it to RGB so we can add coloured rectangles.
markerImage = markerImage.convert('RGB')
# Loop over every image in the pyramid to find matches.
for image in pyramid:
# Get the normalized cross correlation of the template with the image.
grey_image = image.convert('L')
#crossXC = ncc.normxcorr2D(image, template)
crossXC = ncc.normxcorr2D(grey_image, template)
# We loop through this 2D returned array to check for values larger than our threshold.
for y in range(len(crossXC)):
for x in range(len(crossXC[y])):
if crossXC[y][x] > threshold:
matches.append((x*scaleBack,y*scaleBack,scaleBack))
scaleBack *= 1/scale
return matches
def FindTemplate(image, template, threshold):
template = Image.open(template)
#apply a gussian filter on template based on the imgae size to get a better result
template1 = scipy.ndimage.filters.gaussian_filter(template, sigma =3)
#change the template size to given size in the question for original image
#and sclaed version of it based on i
temp_size = (15,int((15*50)/37))
template1 = template.resize(temp_size, Image.BICUBIC)
# apply normalized autocorrelation of template on the image to find matched places
template_finder = ncc.normxcorr2D(image, template1)
template_finder = np.transpose(template_finder)
template_finder = template_finder.reshape((image.size[0], image.size[1]))
image = image.convert('RGB')
#If the array's elements are less than
#threshold they're going to be 0 and else 1
for x in range(0,template_finder.shape[0]):
for y in range(0,template_finder.shape[1]):
if(template_finder[x][y] >= threshold):
template_finder[x][y] = 1
if(template_finder[x][y] < threshold):
template_finder[x][y] = 0
#find the location of elements which are not 0, and draw a rectangle around it based on the template size
for x in range(0,template_finder.shape[0]):
for y in range(0,template_finder.shape[1]):
if(template_finder[x][y] == 1):
x1 = x-int(temp_size[0])
x2 = x+int(temp_size[0])
y1 = y-int(temp_size[1])
y2 = y+int(temp_size[1])
draw = ImageDraw.Draw(image)
draw.line((x1,y1,x2,y1),fill="red",width=2)
del draw
draw = ImageDraw.Draw(image)
draw.line((x1,y1,x1,y2),fill="red",width=2)
del draw
draw = ImageDraw.Draw(image)
draw.line((x1,y2,x2,y2),fill="red",width=2)
del draw
draw = ImageDraw.Draw(image)
draw.line((x2,y2,x2,y1),fill="red",width=2)
del draw
image.save('111.jpg' )
return image.convert('RGB')
def FindTemplate(pyramid, template, threshold):
newTemplateWidth = templateWidth;
newTemplateHeight = templateWidth/template.width*template.height;
if template.width > templateWidth:
template = template.resize((int(newTemplateWidth),int(newTemplateHeight)), Image.BICUBIC)
outputImage = pyramid[0]; # Get the original image, this is the one you draw rectangles on
outputImage = outputImage.convert('RGB')
for index, im in enumerate(pyramid):
nccImage = ncc.normxcorr2D(im, template) # Get the normalized correlation of the template onto the im
rectWidth = newTemplateWidth * (0.75 ** index) # Update template rectangle size
rectHeight = newTemplateHeight * (0.75 ** index) # Update template rectangle size
for row in range(nccImage.shape[0]):
for col in range(nccImage.shape[1]):
if(nccImage[row][col] > threshold):
scaledBackCol = col*(1.333**index)
scaledBackRow = row*(1.333**index)
draw = ImageDraw.Draw(outputImage)
draw.rectangle([scaledBackCol-rectWidth/2, scaledBackRow-rectHeight/2, scaledBackCol+rectWidth/2, scaledBackRow+rectHeight/2], fill=None, outline="red")
del draw
outputImage.show()
def FindTemplate(pyramid, template, threshold):
matches = []
for img in pyramid:
matches.append(ncc.normxcorr2D(img, template))
thresholdMatches = []
#For the matched images in pyramid, diregard the ones
#With small correlations
for match in matches:
thresholdMatches.append(np.where(match >= threshold, 1, 0))
return thresholdMatches
def FindTemplate(pyramid, template, threshold):
# create a new template with width = 15
width = 15
r = template.size[0] / width
height = template.size[1] / r
newTemplate = template.resize((width,height),Image.BICUBIC)
pixelPositions = []
for img in pyramid:
result = ncc.normxcorr2D(img, newTemplate)
# correlation is bigger than the threshold
pixelBiggerThreshold = np.where (result > threshold)
# pixelBiggerThreshold contains two elements, for each x and y
# [0] holds y indices, [1] holds x indices
c = zip(pixelBiggerThreshold[0],pixelBiggerThreshold[1])
# pixelPositions contains all matching points for each image
pixelPositions.append(c)
# make image support RBG color, then we can draw the red retangele on the image
image = pyramid[0].convert('RGB')
for index in range(len(pixelPositions)):
points = pixelPositions[index]
# changes in x, width and y, height
changeInX = newTemplate.size[0]
changeInY = newTemplate.size[1]
# draw red rectangle around the centred of each point
for point in points:
# rescale the poins by the right power of 0.75
xP = point[1] // 0.75 ** index
yP = point[0] // 0.75 ** index
# calculate 4 points around the centre
x2 = xP + changeInX
x1 = xP - changeInX
y1 = yP - changeInY
y2 = yP + changeInY
# draw all 4 lines
draw = ImageDraw.Draw(image)
draw.line((x1,y1,x2,y1),fill="red",width=2)
draw.line((x1,y2,x2,y2),fill="red",width=2)
draw.line((x1,y1,x1,y2),fill="red",width=2)
draw.line((x2,y1,x2,y2),fill="red",width=2)
del draw
image.show()
>>> def FindTemplate(pyramid, template, threshold):
# initialize values
tracker = []
# NCC is expensive, so reduce to a more preferable width
prefered_width = 15
# resize the template with bicubic interpolation to prefered_width
# since not all image are perfect suqare, we need to calcuate how much
# to resize its Y and still retain the image's aspect ratio
scaledown = template.size[0]/prefered_width
template = template.resize((prefered_width, template.size[1]//scaledown), Image.BICUBIC)
# loop through each image in pyramid, compute the NCC
for im in pyramid:
# the array of cross-correlation coefficients, in the range -1.0 to 1.0.
# check whether the threshold is reached
thresholdLevel = np.where(ncc.normxcorr2D(im, template) > threshold)
thresholdLevelat0 = thresholdLevel[0]
thresholdLevelat1 = thresholdLevel[1]
# append the thresholdLvl to the tracker
tracker.append(zip(thresholdLevelat1, thresholdLevelat0))
# If img must support colour. If necessary, you can achieve this with img = img.convert('RGB').
img = pyramid[0].convert('RGB')
# For each pixel at which the normalized correlation result is above
# the threshold, draw lines (forming a rectangle) to mark the boundary of the template
# at that location. All lines should be shown overlayed on the original image.
numofitem = len(tracker)
for pos in range(numofitem):
multiplier = 0.75 ** pos
multiplier2X = 2 * multiplier
# for each item in the array
for thres in tracker[pos]:
# rectangle
x1 = (thres[0]//multiplier) - (template.size[0]//multiplier2X)
y1 = (thres[1]//multiplier) - (template.size[1]//multiplier2X)
x2 = (thres[0]//multiplier) + (template.size[0]//multiplier2X)
y2 = (thres[1]//multiplier) + (template.size[1]//multiplier2X)
# draw read line, calculated rectangle
draw = ImageDraw.Draw(img)
draw.rectangle((x1,y1,x2,y2),outline="red")
del draw
return img
def FindTemplate(pyramid, template, threshold):
desiredWidth = 15
originalHeight = template.size[1]
originalWidth = template.size[0]
ratio = originalWidth / desiredWidth
#Resize the template where width = desiredWidth and height is also reduced to fit the original ratio
template = template.resize((desiredWidth, originalHeight / ratio),
Image.BICUBIC)
#Find all the coordinates in the image that have correlation values with the template > threshold
coordList = []
originalImage = pyramid[0].convert('RGB')
for image in pyramid:
coord = np.where(ncc.normxcorr2D(image, template) > threshold)
x_coord = coord[0]
y_coord = coord[1]
coords = zip(y_coord, x_coord)
coordList.append(coords)
for imagePos in range(len(coordList)):
#Take the power of 0.75 to image position in the pyramidso we adjust according to the size of the picture in our pyramid
scale = 0.75**imagePos
for pixel in coordList[imagePos]:
#Adjust the pixel according to the size of the image in the pyramid
center_x = pixel[0] / scale
center_y = pixel[1] / scale
#Add or subtract a certain offset to the center coordinate so that we get four different coordinates to make a box
offset = 15
left_x = center_x - offset
right_x = center_x + offset
up_y = center_y + offset
bottom_y = center_y - offset
#Draw the lines to specify the pixels on the original image with correlation values > threshold
draw = ImageDraw.Draw(originalImage)
#Draw the horizontal lines
draw.line((left_x, up_y, right_x, up_y), fill="red", width=2)
draw.line((left_x, bottom_y, right_x, bottom_y),
fill="red",
width=2)
#Draw the vertical lines
draw.line((left_x, up_y, left_x, bottom_y), fill="red", width=2)
draw.line((right_x, up_y, right_x, bottom_y), fill="red", width=2)
del draw
return originalImage
def FindTemplate(pyramid, template, threshold):
"Find and mark all locations in the pyramid at which the NCC of the template with the image is above the threshold."
assert (len(pyramid) >
0), "The pyramid must have at least one image element."
template_width = 15
template = Image.open(template)
tx, ty = template.size
t = template.resize((template_width, ty * template_width / tx),
Image.BICUBIC)
im = pyramid[0].copy().convert('RGB')
draw = ImageDraw.Draw(im)
for i, image in enumerate(pyramid):
arr_nmcorrl = ncc.normxcorr2D(image, t)
upscale = (1 /
scale)**i # scale to be multiple regards to original image
height, width = arr_nmcorrl.shape
# To find every pixel that is above the threshold, and output the peaks scaled to the original image
peaks = []
for x in range(width):
for y in range(height):
if (arr_nmcorrl[y][x] > threshold):
peaks.append([x * upscale, y * upscale])
for p in peaks:
x_left = max(p[0] - t.size[0] * upscale, 0)
y_top = max(p[1] - t.size[1] * upscale, 0)
x_right = min(p[0] + t.size[0] * upscale, im.size[0] - 1)
y_bottom = min(p[1] + t.size[1] * upscale, im.size[1] - 1)
# draw a red ractangle box for matchings
drawRedLine(draw, x_left, y_top, x_right, y_top) #top
drawRedLine(draw, x_left, y_top, x_left, y_bottom) #left
drawRedLine(draw, x_left, y_bottom, x_right, y_bottom) #bottom
drawRedLine(draw, x_right, y_top, x_right, y_bottom) #right
del draw
im.show()
return im
# q5
# q6
def FindTemplate(pyramid, template, threshold):
#resize the template to fix width 15px,and scale the height accordingly
template_size = 15
template = Image.open(template)
template = template.resize(
(int(template_size),
int(template.size[1] * template_size / template.size[0])),
Image.BICUBIC)
image = pyramid[0].convert('RGB')
scale = 1
for im in pyramid:
ncc_result = ncc.normxcorr2D(im, template)
#get all the locations that are aboved threshold
above_threshold = np.where(ncc_result >= threshold)
x = above_threshold[1]
y = above_threshold[0]
#adjust template by the right scale factor
backToScale = 1 / scale
xoffset = template.size[0] * backToScale / 2
yoffset = template.size[1] * backToScale / 2
arraySize = len(x)
for i in range(arraySize):
#adjust x and by by the right scale factor
xscaled = x[i] * backToScale
yscaled = y[i] * backToScale
#get the position for top-left,top-right,bottom-left,bottom-right
x1 = xscaled - xoffset
y1 = yscaled - yoffset
x2 = xscaled + xoffset
y2 = yscaled - yoffset
x3 = xscaled + xoffset
y3 = yscaled + yoffset
x4 = xscaled - xoffset
y4 = yscaled + yoffset
#draw line based on the positions
draw = ImageDraw.Draw(image)
draw.line((x1, y1, x2, y2), fill="red", width=2)
draw.line((x2, y2, x3, y3), fill="red", width=2)
draw.line((x3, y3, x4, y4), fill="red", width=2)
draw.line((x4, y4, x1, y1), fill="red", width=2)
scale = scale * 0.75
return image
def FindTemplate(pyramid, template, THRESHOLD):
"""
Finds the correlation between pyramid and template
[I]:
----------
pyramid A list of resized images
template The given template used to match the images on the pyramid
THRESHOLD Is the constant we use to discard small correlations
[O]:
----------
None
"""
match = []
for image in pyramid:
match.append(ncc.normxcorr2D(image, template))
thresholded_match_list = []
for m in match:
thresholded_match_list.append(np.where(m >= THRESHOLD, 1, 0))
return thresholded_match_list
def FindTemplate(pyramid, template, threshold):
'''
Finds and marks all locations in pyramid at which the normalized
cross correlation of the template with the image is above the threshold.
Returns a PIL image of the largest image in the pyramid marked with red
rectangle's corresponding to the locations of template matches
'''
goalWidth = 15
# resize template
ratio = template.size[0] / goalWidth
template = template.resize((goalWidth, template.size[1] // ratio), Image.BICUBIC)
pointLists = []
for image in pyramid:
nccResult = ncc.normxcorr2D(image, template)
aboveThreshold = np.where(nccResult > threshold)
pointLists.append(zip(aboveThreshold[1], aboveThreshold[0]))
convert = pyramid[0].convert('RGB')
for i in range(len(pointLists)):
pointList = pointLists[i]
scaleFactor = 0.75 ** i
for pt in pointList:
ptx = pt[0] // scaleFactor
pty = pt[1] // scaleFactor
adjustx = template.size[0] // (2 * scaleFactor)
adjusty = template.size[1] // (2 * scaleFactor)
x1 = ptx - adjustx
y1 = pty - adjusty
x2 = ptx + adjustx
y2 = pty + adjusty
draw = ImageDraw.Draw(convert)
draw.rectangle([x1,y1,x2,y2], outline="red")
del draw
return convert
def FindTemplate(pyramid, template, threshold):
array = ncc.normxcorr2D(image, template);
matches = list();
for x in range(len(array)):
for y in range(len(array)):
if (array[x][y] > threshold):
matches.append((x,y));
draw = ImageDraw.Draw(pyramid);
for p in matches:
draw.point(p,fill="red");
del draw;
pyramid.show();
#print array;
# go through the array and find all locations above 'threshold'
# for all locations draw them on the original image
return;
def FindTemplate(pyramid, template, threshold):
finalWidth= 15
downsampleScale=template.size[0]/finalWidth
finalHeight=template.size[1]/downsampleScale
template = template.resize((finalWidth, finalHeight), Image.BICUBIC)
#print template.size
img = pyramid[0].convert('RGB')
print 'the length of the pyramid'
print len(pyramid)
for image in pyramid:
NCC=ncc.normxcorr2D(image, template)
print 'NCC value'
print NCC
aboveThresh=np.where(NCC > threshold)
print 'the points above threshold'
print aboveThresh
xlist=aboveThresh[1]
ylist=aboveThresh[0]
print xlist
print ylist
ptx1=template.size[0]/2
pty1=template.size[1]/2
print len(xlist)
for i in range(len(xlist)):
x1=xlist[i]-ptx1
y1=ylist[i]-pty1
x2=xlist[i]+ptx1
y2=ylist[i]+pty1
print x1,x2,y1,y2
draw=ImageDraw.Draw(img)
draw.rectangle((x1,y1,x2,y2),outline="red")
#del draw
img.save('result.png','PNG')
return img
def GetMatchedTemplateCoordinates(pyramid, template, threshold):
x, y = template.size
xyratio = float(y) / float(x)
normtemplate = template.resize(
(TEMPLATE_WIDTH, int(float(TEMPLATE_WIDTH) * (xyratio))),
Image.BICUBIC)
# an array of normalized correlations for each image in the pyramid
correlatedarray = []
# an array of tuples (x,y,index) where index is the index in the pyramid
# this is in order to kow how much to scale the template later
matchingvalues = []
for i in range(0, len(pyramid)):
# get correlation result from template and an image in the pryamid
correlatedarray.append(ncc.normxcorr2D(pyramid[i], normtemplate))
for x in range(0, correlatedarray[i].shape[0]):
for y in range(0, correlatedarray[i].shape[1]):
denormal = correlatedarray[i][x][y]
# if we have value above the threshold, we add that coorinate to the array for later
if correlatedarray[i][x][y] > float(threshold):
scalefactor = 1.0 / pow(.75, i)
matchingvalues.append((int(float(x) * scalefactor),
int(float(y) * scalefactor), i))
return matchingvalues
def FindTemplate(pyramid, template, threshhold):
# maintain a variable so we know by how much to scale the coordinates back.
scaleBack = 1
# this marker will be the image where we draw the rectangles of matches we find.
marker = pyramid[0]
marker = marker.convert('RGB')
# loop over every image in the pyramid to find matches.
for image in pyramid:
# Get the normalized cross correlation of the template with the image.
crossXC = ncc.normxcorr2D(image, template)
# loop through this 2D returned array to check for values larger than our threshold.
for y in range(len(crossXC)):
for x in range(len(crossXC[y])):
if crossXC[y][x] > threshhold:
# find scaled coordinate values
draw = ImageDraw.Draw(marker)
x1 = x*scaleBack + template.size[0]
x2 = x*scaleBack - template.size[0]
y1 = y*scaleBack - template.size[1]
y2 = y*scaleBack + template.size[1]
# draw a red rectangle around
draw.line((x1,y1,x2,y1),fill="red",width=2)
draw.line((x1,y2,x2,y2),fill="red",width=2)
draw.line((x1,y1,x1,y2),fill="red",width=2)
draw.line((x2,y1,x2,y2),fill="red",width=2)
del draw
# scaleback is determined the inverse of 0.75, our original scale value - (0.75)^(-1) = 1.33333333
scaleBack *= 1.33333
return marker
def FindTemplate(pyramid, template, threshold):
#reduce template to 15 pixels/define a constant for template width
size_template = 15.0
template = Image.open("/Users/leticianakajima/PycharmProjects/a2/" +
template + ".jpg")
width, height = template.size
scaling_factor = size_template / min(width, height)
new_height = scaling_factor * height
new_width = scaling_factor * width
new_template = template.resize((int(new_width), int(new_height)),
PIL.Image.BICUBIC)
matches_array = []
#ncc-ing the template and all the pyramid images
for x in pyramid:
matches = np.where(ncc.normxcorr2D(x, new_template) > threshold)
x = matches[1]
y = matches[0]
#adding the matches to this array
matches_array.append(zip(x, y))
#passing it into a helping draw function
img = HelperDrawer(pyramid[0], matches_array)
img.save('faces_recon.png', 'PNG')
def FindTemplate(pyramid, template, threshhold):
# maintain a variable so we know by how much to scale the coordinates back.
scaleBack = 1
# this marker will be the image where we draw the rectangles of matches we find.
marker = pyramid[0]
marker = marker.convert('RGB')
# loop over every image in the pyramid to find matches.
for image in pyramid:
# Get the normalized cross correlation of the template with the image.
crossXC = ncc.normxcorr2D(image, template)
# loop through this 2D returned array to check for values larger than our threshold.
for y in range(len(crossXC)):
for x in range(len(crossXC[y])):
if crossXC[y][x] > threshhold:
# find scaled coordinate values
draw = ImageDraw.Draw(marker)
x1 = x * scaleBack + template.size[0]
x2 = x * scaleBack - template.size[0]
y1 = y * scaleBack - template.size[1]
y2 = y * scaleBack + template.size[1]
# draw a red rectangle around
draw.line((x1, y1, x2, y1), fill="red", width=2)
draw.line((x1, y2, x2, y2), fill="red", width=2)
draw.line((x1, y1, x1, y2), fill="red", width=2)
draw.line((x2, y1, x2, y2), fill="red", width=2)
del draw
# scaleback is determined the inverse of 0.75, our original scale value - (0.75)^(-1) = 1.33333333
scaleBack *= 1.33333
return marker
def FindTemplate(pyramid, template, threshold):
# Set the TEMPLATE_WIDTH constant (=15 as given in specifications)
TEMPLATE_WIDTH = 15
# Set the output image to be the first image of the pyramid (which has the size as the original image without any scaling)
image = pyramid[0]
# Convert the image from grayscale to color in order to see the face detection rectangle in colors (="red" rectangle as required)
image = image.convert('RGB')
# initialize the draw object on the given image
draw = ImageDraw.Draw(image)
# Open the image, here the template, so that can be accessed and used while performing the Normalized Cross-Correlation
with Image.open(template) as template:
# Set x to be the temlate's width and y to be the template's height
x, y = template.size
# Set the template width to TEMPLATE_WIDTH (= 15, as asked)
x = TEMPLATE_WIDTH
# Set the template height to a value that keeps the original ratio between width and height
y = (float(TEMPLATE_WIDTH) / template.width) * template.height
# Resize Template to width = TEMPLATE_WIDTH (= 15, as asked), and keep the ratio for the height
template = template.resize((int(x), int(y)), Image.BICUBIC)
# Scan through every photo in the pyramid using the template
for photo_index in range(0, len(pyramid)):
# Perform the Normalized Cross-Correcation on the given image using the given template
output = ncc.normxcorr2D(pyramid[photo_index], template)
# Save the Matrix where the normalized correlation result is above the threshold
comparison = np.where(output > threshold)
# Save the row of the Matrix where the normalized correlation result is above the threshold
row = comparison[1]
# Save the column of the Matrix where the normalized correlation result is above the threshold
column = comparison[0]
# Loop through every element in the row (or column, since they are equal in length)
for i, j in enumerate(row):
# Rescale the width of the red rectangle after each iteration using the given ratio (template.width * 1/0.75^i)
rescaled_x = x * 1 / (0.75**photo_index)
# Rescale the height of the red rectangle after each iteration using the given ratio (template.height * 1/0.75^i)
rescaled_y = y * 1 / (0.75**photo_index)
# Save the row element of the current iteration in M_x (name symbolizes Matrix x axis)
M_x = j
# Save the column element of the current iteration in M_x (name symbolizes Matrix y axis)
M_y = column[i]
# THE FACE DETECTION RECTANGLE BOX
# Bottom line
x1 = M_x * 1 / (
0.75**photo_index
) - rescaled_x / 2 # Rescale the image point x to match x size of the output image
y1 = M_y * 1 / (
0.75**photo_index
) + rescaled_y / 2 # Rescale the image point y to match y size of the output image
x2 = M_x * 1 / (
0.75**photo_index
) + rescaled_x / 2 # Rescale the image point x to match x size of the output image
y2 = M_y * 1 / (
0.75**photo_index
) + rescaled_y / 2 # Rescale the image point y to match y size of the output image
draw.line(
(x1, y1, x2, y2), fill="red", width=2
) # Draw the bottom line of the rectangle with width = 2 (colored in red)
# Right line
x1 = M_x * 1 / (0.75**photo_index) + rescaled_x / 2
y1 = M_y * 1 / (0.75**photo_index) + rescaled_y / 2
x2 = M_x * 1 / (0.75**photo_index) + rescaled_x / 2
y2 = M_y * 1 / (0.75**photo_index) - rescaled_y / 2
draw.line(
(x1, y1, x2, y2), fill="red", width=2
) # Draw the right line of the rectangle with width = 2 (colored in red)
# Top line
x1 = M_x * 1 / (0.75**photo_index) - rescaled_x / 2
y1 = M_y * 1 / (0.75**photo_index) - rescaled_y / 2
x2 = M_x * 1 / (0.75**photo_index) + rescaled_x / 2
y2 = M_y * 1 / (0.75**photo_index) - rescaled_y / 2
draw.line(
(x1, y1, x2, y2), fill="red", width=2
) # Draw the top line of the rectangle with width = 2 (colored in red)
# Left Line
x1 = M_x * 1 / (0.75**photo_index) - rescaled_x / 2
y1 = M_y * 1 / (0.75**photo_index) + rescaled_y / 2
x2 = M_x * 1 / (0.75**photo_index) - rescaled_x / 2
y2 = M_y * 1 / (0.75**photo_index) - rescaled_y / 2
draw.line(
(x1, y1, x2, y2), fill="red", width=2
) # Draw the bottom line of the rectangle with width = 2 (colored in red).
# After drawing this last line (left line), these four lines (bottom, right, top, and left) create a red rectangle
# Remove the binding of the draw from the local namespace
del draw
# Return the final image which contains red rectangles on the detected faces
return image
from PIL import Image, ImageDraw
import numpy as np
import math
from scipy import signal
import ncc
scale = 0.75
threshold = 0.55
pyramid = a2.MakePyramid("./faces/judybats.jpg", 20)
template = "./faces/template.jpg"
template_width = 15
t = Image.open(template).resize((template_width, template_width),
Image.BICUBIC)
for i, image in enumerate(pyramid):
arr_nmcorrl = ncc.normxcorr2D(image, t)
upscale = (1 / scale)**i # scale to be multiple regards to original image
#height, width = arr_nmcorrl.shape
width, height = arr_nmcorrl.shape[::-1]
print width, height
#print range(width), range(height)
# To find every pixel that is above the threshold, and output the peaks scale to the original image
peaks = []
for x in range(width):
for y in range(height):
if (arr_nmcorrl[y][x] > threshold):
peaks.append([x * upscale, y * upscale])
print peaks
import a2
threshold = 0.571