def filterPoints(method, filters, alpha, K, source, name):
fname1 = os.path.join(source, name)
img = sp.ndimage.imread(fname1)
if img.ndim == 3: #only use color images
pixelResponses = np.zeros((alpha, 3 * len(filters)))
imageResponse = extractFilterResponses(img, filters) #apply filters
#Get alpha points for each image
if method == 'Random':
points = getRandomPoints(img, alpha)
elif method == 'Harris':
points = getHarrisPoints(img, alpha, K)
#Extract filter responses at points
for row in range(0, alpha):
x, y = points[row, :]
pixelResponses[row] = imageResponse[x, y, :]
return pixelResponses
def getVisualWords(I, dictionary, filterBank):
print("Getting Visual Words")
imgResponse = extractFilterResponses(
I, filterBank) #apply filters, (H,W,3n) size
H, W, filt = imgResponse.shape
imgResponse = imgResponse.reshape((H * W, filt)) #each row is a point
dist = sp.spatial.distance.cdist(imgResponse,
dictionary,
metric='euclidean')
wordMap = dist.argmin(
axis=1
) #find minimum of each row aka what cluster is closest, returns a shape (H*w,1)
wordMap = wordMap.reshape((H, W)) #2D
return wordMap
import os
from createFilterBank import createFilterBank
from extractFilterResponses import extractFilterResponses
import scipy as sp
fname = '../data/campus/sun_ahprylpgnmgqiyuz.jpg'
img = sp.ndimage.imread(fname)
#testing for Q1.1 Filter responses
filters = createFilterBank()
Im = extractFilterResponses(img,filters)
image1 = Im[:,:,0:3] #guassian 0,3,6,9,12,|| 15,18,21,24,27 || ,30,33,36,39,42, || 45,48,51,54
image2 = Im[:,:,30:33]
image3 = Im[:,:,45:48]
sp.misc.imsave('../og.png', img)
sp.misc.imsave('../filter1.png', image1)
sp.misc.imsave('../filter2.png', image2)
sp.misc.imsave('../filter3.png', image3)