def __getOnePictureRegion(self,img):
'''
get region of the picture
return:
the boundingbox list of the image
and it save as ((xmin,ymin),(xmax,ymax))
'''
if img!=None:
h_cof = float(img.shape[0])/self.height
w_cof = float(img.shape[1])/self.width
#resize the img to standard scale
img = cv.resize(img,(self.height,self.width))
#blur the image to remove the noise
img = cv.GaussianBlur(img,(5,5),0.8,0.8,0)
#get the edge set of the image
e_set = []
for i in range(0,self.height):
for j in range(0,self.width):
ij_edge=[]
if j<self.width-1:
similar = abs(img.item(i,j)-img.item(i,j+1))
ij_edge.append((i*self.width+j+1,similar))
if i<self.height-1:
similar = abs(img.item(i,j)-img.item(i+1,j))
ij_edge.append(((i+1)*self.width+j,similar))
if j<self.width-1 and i<self.height-1:
similar = abs(img.item(i,j)-img.item(i+1,j+1))
ij_edge.append(((i+1)*self.width+(j+1),similar))
e_set.append(ij_edge)
#do segment
img_seg = segmentor.grap_base_seg(e_set,self.k)
#return result in boundingbox list
bbs = {}
for i in range(0,len(img_seg)):
raw = i//self.width
column = i-raw*self.width
region = disjoin_set.find(img_seg[i])
if bbs.has_key(region):
if bbs[region][0] > column:
bbs[region][0] = column
if bbs[region][2] < column:
bbs[region][2] = column
if bbs[region][1] > raw:
bbs[region][1] = raw
if bbs[region][3] < raw:
bbs[region][3] = raw
else:
bbs[region] = [column,raw,column,raw]
regions = []
filter_size = self.height*self.width*self.thresh
for key in bbs:
if (bbs[key][2]-bbs[key][0])\
*(bbs[key][3]-bbs[key][1])>=filter_size:
regions.append(
[int(bbs[key][0]*w_cof),\
int(bbs[key][1]*h_cof),\
int(bbs[key][2]*w_cof),\
int(bbs[key][3]*h_cof)]
)
return regions
def main(data):
forest = []
for item in data:
forest.append(node(item))
disjoin_set.union(forest[1],forest[2])
disjoin_set.union(forest[3],forest[4])
disjoin_set.union(forest[1],forest[4])
for item in forest:
if item==disjoin_set.find(item):
print "item is:"+str(item)+" value is:"+str(item.value)+" parent is:"+str(item.parent)
def grap_base_seg(e_set,k):
'''
Here we will implimention the graph based iamge segment method
which proposed by Pedro F.Felzenszwalb and Daniel P.Huttenlocher
in the paper "Efficient Graph-Based Image Segmentation"
This method based on Kruskal's algorithm and use the disjoin set
as the data structure
The interpretation of arguments:
e_set : is the list of edge's weight, and it saved in a
symmetric matrix. And the index is the position of pixel
which ordered by column first, raw second.As the matrix will
be very sparse, so here use link chain to store
k : is the coefficent multiplier
The result will stored in the disjoin set
'''
#Zero: order the edge in e_set by weight in Ascending order
ord_e = []
for i in range(0,len(e_set)):
for j in range(0,len(e_set[i])):
ord_e.append(edge_node(i,e_set[i][j][0],e_set[i][j][1]))
ord_e.sort(cmp_e)
#First: initialize the segment,v_set will maintains the result
v_set = []
for raw in range(0,len(e_set)):
v_set.append(vertex_node(0))
#Second, split loop, for each edge do the follow
for e in ord_e:
region1 = disjoin_set.find(v_set[e.v1])
region2 = disjoin_set.find(v_set[e.v2])
if region1!=region2:
Int1 = region1.value + k/region1.coefficient
Int2 = region2.value + k/region2.coefficient
Int = Int1
if Int1 > Int2:
Int = Int2
if e.w <= Int: #then the two region can be merged
region = disjoin_set.union(region1,region2)
region.value = e.w #the max w in MST of region
region.coefficient = region1.coefficient + region2.coefficient
else:
continue
#Third, return the result
return v_set
def grap_base_seg(e_set, k):
"""
Here we will implimention the graph based iamge segment method
which proposed by Pedro F.Felzenszwalb and Daniel P.Huttenlocher
in the paper "Efficient Graph-Based Image Segmentation"
This method based on Kruskal's algorithm and use the disjoin set
as the data structure
The interpretation of arguments:
e_set : is the list of edge's weight, and it saved in a
symmetric matrix. And the index is the position of pixel
which ordered by column first, raw second.As the matrix will
be very sparse, so here use link chain to store
k : is the coefficent multiplier
The result will stored in the disjoin set
"""
# Zero: order the edge in e_set by weight in Ascending order
ord_e = []
for i in range(0, len(e_set)):
for j in range(0, len(e_set[i])):
ord_e.append(edge_node(i, e_set[i][j][0], e_set[i][j][1]))
ord_e.sort(cmp_e)
# First: initialize the segment,v_set will maintains the result
v_set = []
for raw in range(0, len(e_set)):
v_set.append(vertex_node(0))
# Second, split loop, for each edge do the follow
for e in ord_e:
region1 = disjoin_set.find(v_set[e.v1])
region2 = disjoin_set.find(v_set[e.v2])
if region1 != region2:
Int1 = region1.value + k / region1.coefficient
Int2 = region2.value + k / region2.coefficient
Int = Int1
if Int1 > Int2:
Int = Int2
if e.w <= Int: # then the two region can be merged
region = disjoin_set.union(region1, region2)
region.value = e.w # the max w in MST of region
region.coefficient = region1.coefficient + region2.coefficient
else:
continue
# Third, return the result
return v_set
def main(data):
forest = []
for item in data:
forest.append(node(item))
disjoin_set.union(forest[1], forest[2])
disjoin_set.union(forest[3], forest[4])
disjoin_set.union(forest[1], forest[4])
for item in forest:
if item == disjoin_set.find(item):
print "item is:" + str(item) + " value is:" + str(
item.value) + " parent is:" + str(item.parent)
def initial_set(img, initial_regions, bins, similar_coef, similar_set,
region_set, bb_set):
width = img.shape[1]
height = img.shape[0]
img_size = img.size
for i in range(0, len(initial_regions)):
pixel = (i // width, i - (i // width) * width)
color = img[pixel[0], pixel[1]]
region = disjoin_set.find(initial_regions[i])
if region_set.has_key(region):
region_set[region].add_p(pixel, color)
else:
region_set[region] = region_node()
region_set[region].add_p(pixel, color)
for region in region_set:
his = region_similar.histogram(region_set[region].region_value,
bins)
region_set[region].colour_his = his / his.sum()
bb_set.append(region_set[region].bounding_box)
for i in range(0, len(initial_regions)):
raw = i // width
column = i - raw * width
if column < width - 1:
j = i + 1
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1 != region2:
selective_search.make_pair(img_size, similar_set,
region_set, similar_coef,
region1, region2)
if raw < height - 1:
j = (raw + 1) * width + column
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1 != region2:
selective_search.make_pair(img_size, similar_set,
region_set, similar_coef,
region1, region2)
if raw < height - 1 and column < width - 1:
j = (raw + 1) * width + (column + 1)
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1 != region2:
selective_search.make_pair(img_size, similar_set,
region_set, similar_coef,
region1, region2)
def main(img_path, thresh):
#load the image
raw_img = cv.imread(img_path)
img = cv.cvtColor(raw_img, cv.COLOR_RGB2GRAY) #use gray image
if img == None:
sys.stderr.write("error happened in Loading img!")
return
cv.namedWindow('image', cv.WINDOW_NORMAL) # create a can resize window
cv.imshow('image', img) #draw image in window named "image"
cv.waitKey(0) #until a key stroke, it will continue
#resize the iamge
img = cv.resize(img, (128, 128))
raw_img = cv.resize(raw_img, (128, 128))
#blur to remove the noise
img = cv.GaussianBlur(img, (5, 5), 0.8, 0.8, 0)
#Get the edge set of the image
e_set = []
height = img.shape[0]
width = img.shape[1]
for i in range(0, height):
for j in range(0, width):
ij_edge = []
if j < width - 1:
similar = abs(img.item(i, j) - img.item(i, j + 1))
#similar = pixel_similar.normalize_dif((i,j),(i,j+1),img,5)
ij_edge.append((i * width + j + 1, similar))
if i < height - 1:
similar = abs(img.item(i, j) - img.item(i + 1, j))
#similar = pixel_similar.normalize_dif((i,j),(i+1,j),img,5)
ij_edge.append(((i + 1) * width + j, similar))
if j < width - 1 and i < height - 1:
similar = abs(img.item(i, j) - img.item(i + 1, j + 1))
#similar = pixel_similar.normalize_dif((i,j),(i+1,j+1),img,5)
ij_edge.append(((i + 1) * width + (j + 1), similar))
e_set.append(ij_edge)
print "The number of vertex is: " + str(len(e_set))
#do segment
k = 350 #(which is good for image with size: 128*128)
img_seg = segmentor.grap_base_seg(e_set, k)
print len(img_seg)
count = 0
for item in img_seg:
if item == item.parent:
count += 1
print "region number is: " + str(count)
#draw the result
regions = {}
'''
for i in range(0,len(img_seg)):
raw = i//width
column = i-(raw*width)
region = disjoin_set.find(img_seg[i])
if regions.has_key(region):
if regions[region][0]>raw:
regions[region][0]=raw
elif regions[region][1]<raw:
regions[region][1] = raw
else:
pass
if regions[region][2]>column:
regions[region][2] = column
elif regions[region][3]<column:
regions[region][3] = column
else:
pass
else:
regions[region]=[raw,raw,column,column]
'''
for i in range(0, len(img_seg)):
raw = i // width
column = i - (raw * width)
region = disjoin_set.find(img_seg[i])
if regions.has_key(region):
regions[region].append((raw, column))
else:
regions[region] = [(raw, column)]
pixel_num = 0
for region in regions:
color = [
random.randint(0, 255),
random.randint(0, 255),
random.randint(0, 255)
]
pixel_num += len(regions[region])
if len(regions[region]) > thresh:
for p in regions[region]:
raw_img[p[0], p[1]] = color
print "The number of pixel in region is: " + str(pixel_num)
cv.imshow('res', raw_img)
cv.waitKey(0)
cv.destroyAllWindows()
'''
def initial_set(img,initial_regions,bins,similar_coef,similar_set,region_set,bb_set):
width = img.shape[1]
height = img.shape[0]
img_size = img.size
for i in range(0,len(initial_regions)):
pixel = (i//width,i-(i//width)*width)
color = img[pixel[0],pixel[1]]
region = disjoin_set.find(initial_regions[i])
if region_set.has_key(region):
region_set[region].add_p(pixel,color)
else:
region_set[region] = region_node()
region_set[region].add_p(pixel,color)
for region in region_set:
his = region_similar.histogram(
region_set[region].region_value,
bins)
region_set[region].colour_his = his/his.sum()
bb_set.append(region_set[region].bounding_box)
for i in range(0,len(initial_regions)):
raw = i//width
column = i-raw*width
if column < width-1:
j = i+1
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1!=region2:
selective_search.make_pair(
img_size,
similar_set,
region_set,
similar_coef,
region1,
region2
)
if raw < height-1:
j = (raw+1)*width+column
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1!=region2:
selective_search.make_pair(
img_size,
similar_set,
region_set,
similar_coef,
region1,
region2
)
if raw < height-1 and column < width-1:
j = (raw+1)*width+(column+1)
region1 = disjoin_set.find(initial_regions[i])
region2 = disjoin_set.find(initial_regions[j])
if region1!=region2:
selective_search.make_pair(
img_size,
similar_set,
region_set,
similar_coef,
region1,
region2
)
def __getOnePictureRegion(self, img):
'''
get region of the picture
return:
the boundingbox list of the image
and it save as ((xmin,ymin),(xmax,ymax))
'''
if img != None:
h_cof = float(img.shape[0]) / self.height
w_cof = float(img.shape[1]) / self.width
#resize the img to standard scale
img = cv.resize(img, (self.height, self.width))
#blur the image to remove the noise
img = cv.GaussianBlur(img, (5, 5), 0.8, 0.8, 0)
#get the edge set of the image
e_set = []
for i in range(0, self.height):
for j in range(0, self.width):
ij_edge = []
if j < self.width - 1:
similar = abs(img.item(i, j) - img.item(i, j + 1))
ij_edge.append((i * self.width + j + 1, similar))
if i < self.height - 1:
similar = abs(img.item(i, j) - img.item(i + 1, j))
ij_edge.append(((i + 1) * self.width + j, similar))
if j < self.width - 1 and i < self.height - 1:
similar = abs(img.item(i, j) - img.item(i + 1, j + 1))
ij_edge.append(
((i + 1) * self.width + (j + 1), similar))
e_set.append(ij_edge)
#do segment
img_seg = segmentor.grap_base_seg(e_set, self.k)
#return result in boundingbox list
bbs = {}
for i in range(0, len(img_seg)):
raw = i // self.width
column = i - raw * self.width
region = disjoin_set.find(img_seg[i])
if bbs.has_key(region):
if bbs[region][0] > column:
bbs[region][0] = column
if bbs[region][2] < column:
bbs[region][2] = column
if bbs[region][1] > raw:
bbs[region][1] = raw
if bbs[region][3] < raw:
bbs[region][3] = raw
else:
bbs[region] = [column, raw, column, raw]
regions = []
filter_size = self.height * self.width * self.thresh
for key in bbs:
if (bbs[key][2]-bbs[key][0])\
*(bbs[key][3]-bbs[key][1])>=filter_size:
regions.append(
[int(bbs[key][0]*w_cof),\
int(bbs[key][1]*h_cof),\
int(bbs[key][2]*w_cof),\
int(bbs[key][3]*h_cof)]
)
return regions
def main(img_path,thresh):
#load the image
raw_img = cv.imread(img_path)
img = cv.cvtColor(raw_img,cv.COLOR_RGB2GRAY) #use gray image
if img==None:
sys.stderr.write("error happened in Loading img!")
return
cv.namedWindow('image',cv.WINDOW_NORMAL) # create a can resize window
cv.imshow('image',img) #draw image in window named "image"
cv.waitKey(0) #until a key stroke, it will continue
#resize the iamge
img = cv.resize(img,(128,128))
raw_img = cv.resize(raw_img,(128,128))
#blur to remove the noise
img = cv.GaussianBlur(img,(5,5),0.8,0.8,0)
#Get the edge set of the image
e_set=[]
height = img.shape[0]
width = img.shape[1]
for i in range(0,height):
for j in range(0,width):
ij_edge=[]
if j<width-1:
similar = abs(img.item(i,j)-img.item(i,j+1))
#similar = pixel_similar.normalize_dif((i,j),(i,j+1),img,5)
ij_edge.append((i*width+j+1,similar))
if i<height-1:
similar = abs(img.item(i,j)-img.item(i+1,j))
#similar = pixel_similar.normalize_dif((i,j),(i+1,j),img,5)
ij_edge.append(((i+1)*width+j,similar))
if j<width-1 and i<height-1:
similar = abs(img.item(i,j)-img.item(i+1,j+1))
#similar = pixel_similar.normalize_dif((i,j),(i+1,j+1),img,5)
ij_edge.append(((i+1)*width+(j+1),similar))
e_set.append(ij_edge)
print "The number of vertex is: "+str(len(e_set))
#do segment
k=350 #(which is good for image with size: 128*128)
img_seg = segmentor.grap_base_seg(e_set,k)
print len(img_seg)
count=0
for item in img_seg:
if item == item.parent:
count+=1
print "region number is: "+str(count)
#draw the result
regions = {}
'''
for i in range(0,len(img_seg)):
raw = i//width
column = i-(raw*width)
region = disjoin_set.find(img_seg[i])
if regions.has_key(region):
if regions[region][0]>raw:
regions[region][0]=raw
elif regions[region][1]<raw:
regions[region][1] = raw
else:
pass
if regions[region][2]>column:
regions[region][2] = column
elif regions[region][3]<column:
regions[region][3] = column
else:
pass
else:
regions[region]=[raw,raw,column,column]
'''
for i in range(0,len(img_seg)):
raw = i//width
column = i-(raw*width)
region = disjoin_set.find(img_seg[i])
if regions.has_key(region):
regions[region].append((raw,column))
else:
regions[region]=[(raw,column)]
pixel_num = 0
for region in regions:
color=[random.randint(0,255),random.randint(0,255),random.randint(0,255)]
pixel_num += len(regions[region])
if len(regions[region]) > thresh:
for p in regions[region]:
raw_img[p[0],p[1]] = color
print "The number of pixel in region is: "+str(pixel_num)
cv.imshow('res',raw_img)
cv.waitKey(0)
cv.destroyAllWindows()
'''
