key_frames_0.2.py

#get_key_frames

import numpy as np
import matplotlib.pyplot as plt
import cv2
import pandas as pd
import numpy as np

#LBP FEATURE EXTRACTION

from skimage.transform import rotate
from skimage.feature import local_binary_pattern
from skimage import data
from skimage.color import label2rgb

#get LPBH from image
def lbp_histogram(img,radius):
    #img = color.rgb2gray(color_image)
    patterns = local_binary_pattern(img, 8, 1)
    hist, _ = np.histogram(patterns, bins=np.arange(radius**8 + 1), density=True)
    return hist

def lbp_one(im,radius):
 #radius = 2
 n_points = 8 * radius

 lbp = local_binary_pattern(im, n_points, radius)
 lbph=lbp_histogram(im,radius)
 return lbph

def lbp(im,radius):
 f=[]
 for i in range (len(im)):
  f.append(lbp_one(im[i]),radius)
 
 #print(np.array(f))
 print('LBP feature extracted')
 return avg(np.array(f))

##chiSquared Distance
def chiSquared(p,q):
    return 0.5*np.sum((p-q)**2/(p+q+1e-6))

def divide_image(img,n,m):
 
 slice_total=[]
 # List of n  image slices
 sliced = np.split(img,n,axis=0)

 # List of m lists of n image blocks
 blocks = [np.split(img_slice,m,axis=1) for img_slice in sliced]
 
 for i in blocks:
   for j in i:
     slice_total.append(j)

 return np.array(slice_total)

def get_key_frames_1(img_seq,show_plot=False,N=11,M=10,p_factor=0.5,block_size=(6),lbp_r=2,N_frames=20,constant=False):
 #devide images into blocks
 #img_seq=images_dummy_vid
  max_ind=len(img_seq)
  interval=int ( (N+1)/2 )

  img_blocks=[]
  for i in img_seq:
    img_blocks.append(divide_image((i),block_size,block_size))
  img_blocks=np.array(img_blocks)
  
  #get LBP for each block
  lbp_image=[]
  for i in img_blocks:
    lbp_block=[]
    for j in i:
      lbp_block.append(lbp_one(j,lbp_r))
    lbp_image.append(lbp_block)
  lbp_image=np.array(lbp_image) 

  #calculate diff from HF and TF value for each block
  diff_total=[]
 
  for i in range(interval):
   diff_total.append(np.zeros((block_size*block_size)))

  for i in range(interval,max_ind-interval):
    sf=lbp_image[i-interval]
    ef=lbp_image[i+interval]
    cf=lbp_image[i]
    diff=[]
    for j in range(len(cf)):
      diff.append(chiSquared(cf[j],(sf[j]+ef[j])/2))
    diff_total.append(diff)
 
  for i in range(max_ind-interval,max_ind):
   diff_total.append(np.zeros((block_size*block_size)))
  
  diff_total=np.array(diff_total)

  f=[]
  for i in diff_total:
   f.append(np.sum(np.sort(i)[M:])/M)
  f=np.array(f)
 
  c=[]
  for i in range(N+1):
   c.append(0)

  for i in range(N+1,max_ind-N-1):
      c.append(f[i]-(f[i-interval]+f[i+interval])/2)  

  for i in range(max_ind-N-1,max_ind):
   c.append(0)

  c=np.array(c)

  th=np.mean(c) +( p_factor *(np.max(c)-np.mean(c)))
  
  th_line=th*np.ones(len(c)) 
  
  c = c.clip(min=0)

  if(show_plot):
   fig = plt.figure(figsize=(12,9))
   ax = fig.subplots(1) 
   ax.plot(c, linestyle='--',color='r')
   ax.plot(th_line,linestyle='--')
   plt.xlabel('Frame Number',fontsize=18)
   plt.ylabel('Difference Measure',fontsize=18)
   #ax.set_yticks([])
   #ax.set_xticks([])
   ax.grid(True)
   plt.show()
  
  if(constant):
   candidate=(sorted(range(len((c-th))), key=lambda i: (c-th)[i])[-N_frames:])
   return np.sort(candidate)
  else:
   ret_f=[]    
   for i in range(len(c)):
       if(c[i]>=th):
        ret_f.append(i)
   return np.array(ret_f)