# C14399846 DIT COMPUTER SCIENCE

# (C) OLEG PETCOV

# import the necessary packages:

from __future__ import division

from matplotlib import pyplot as plt

from matplotlib import image as image

import numpy as np

import os, errno

import sys

import math

import cv2

import imutils

import easygui

import time

'''

HOW TO USE:

NEED TO PIP INSTALL:

[imutils, easygui]

pip install imutils

pip install easygui

The user will be able to choose a video file from directory

A folder 'cleaned' is created if not exists already, from launch place of script.

Bools:

[makingVideo]

Set to 'True' if making a video file

Set to 'False' if making individual frames

[forceFrame]

Set to 'True' if starting from a specific frame,

and ending after 'frameStop' amount of frames

Set to 'False' if creating from frame 0,

and computing the whole video

'''

'''

THE PROGRAM:

The purpose of this python script is to improve the quality of old film-based clips.

It will employ a range of functions, to clean up a video frame-by-frame.

###############################################################################################################

1)

Initially the user will choose a video file from directory.

###############################################################################################################

2)

The first frame will be used to take out any watermarks within the image,

as long as the first frame is a largely black background image, with the watermark overlayed onto it.

This is done through 'inpaint()' and an inverse mask.

Reset video to frame 0 again

###############################################################################################################

3)

The whole video is parsed to get rid of the watermark,

and converted to gray.

This is useful for a few reasons.

The main one is for processing using OpenCV functions,

which require a gray / 2d channeled image.

The other reason is performance.

If you attempt to run some functions (like fastNLDenoise),

You will hit HEAVY bottlenecks because the image is passed as a BGR, multi-channeled image.

This is far harder to compute when using the fastNLDenoise for gray images.

The computed image is stored in an array 'Image_data'.

###############################################################################################################

4)

A folder 'cleaned' is created, if it doesn't already exist.

This is used to store processed images and video files.

The situational Bools are set here:

If you want to process video, set Bool 'makingVideo' to True

If you want to process frame-by-frame images, set 'MakingVideo' to False

If you'd like to process only specific frames, e.g 105 to 150, set Bool 'forceFrame' to True

Else, set it to False in order to process the entire video

frame details are initialised:

- totalFrames: length of the array 'Image_data'

- startFrame: starting at this frame, default is 0, for beginnning frame

- stopFrame: stop running after this many frames

- endFrame: last frame, (totalFrames - 1)

- frameNum: counting iterations of frames

###############################################################################################################

5)

ALGO BREAKDOWN

outputs percentage as you run the program.

The percentage code is not mine, highlighted in the code itself.

An image is grabbed from 'Image_data'

###############################################################################################################

i)

Attributes: [bilat, sharpenBil]

It is passed into a bilateral filter and a custom sharpen filter.

This is done to preserve edges and details,

and to not push any actors too heavily into the foreground,

as can be seen in the walls if the sharpen is too heavy.

Sharpen helps to 'pop' details after bilateral is applied.

###############################################################################################################

ii)

Attributes: [dilated,eroded,dilatedFinal]

A dilate, erode, dilate.

Used to smooth out details after the sharpen,

While (lightly) filling in some of the smaller blotches.

They make use of a 5x5 Matrix kernel, and a 3x3 Matrix kernel.

dilates use 3x3 [smallMorph], erode uses [largeMorph]

###############################################################################################################

iii)

Attributes: [fastDen, sharpNoise]

Removes noise from bilateral blur, sharpen and morphs.

Quick sharpen to preserve edges again.

A very light denoise pass here,

to preserve detail for lower functions.

Also to improve performance.

Re-Sharpen to preserve feeature erosion form smoothing

###############################################################################################################

iv)

Attributes: [sharpDil,sharpEro,sharpDilFinal]

Fills in more minor blotches,

and helps to smooth wall and background surfaces

Unfortunately does 'brighten' some whitepoints in frames.

This is due to the double dilates, in both parts of the algo.

A custom equalise would be nice here.

Clahe and the regular equalise fucntion return a non-ideal image.

###############################################################################################################

v)

Attributes: [gBlurred,denoisedGray]

Final blur (Gaussian) and heaviest denoise is applied here.

Edge preservation is less important here, as compared to initial image,

where more preservation is ideal, when you need to do many passes of smoothing, etc.

Detail is lost in denser areas.

Look at cloak / blanket in fight scene, and on stairs beside Zorro.

The top of them is lightly melded together.

Although, walls and background are a lot cleaner now.

The candles are a bit clearer too.

###############################################################################################################

vi)

Attributes: [output,frameNum]

Converts back to BGR for correct video format.

Adds frame to videofile / creates imagefile in specified directory

[frameNum] used to iterate through the array

###############################################################################################################

vii)

Attributes: [time]

Outputs the time to compute algo in seconds

###############################################################################################################

'''

'''

A nifty tool to use for easier comparison is 'combineVideos.py'

This script will combine two video files into one stacked video.

Input the video filenames inside the sscript, no arguments used.

Unfortunately they are a bit 'squashed', because of the stacking,

But it's better than nothing.

It's fairly fast to run too.

Note: **** They must be the same resolution ****

'''

'''

Look into images: ARR.png, ssim.png, mean.png, and mean_inv.png (This one is inversed and rotated)

The large spikes are changes in frames, which can denote a change in scene

One anomaly is towards the end, where the group is running.

This scene causes a lot of flux.

Attaching folder of imagse of the changing scenes.

'''

# 1)

# Opening an image using a File Open dialog:

videofile = easygui.fileopenbox()

# NOT USED, IT MESSES UP WALL SHADOWS

#clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))

# Sharpening kernels

kernelSharp = np.array( [[ 0, -1, 0], [ -1, 5, -1], [ 0, -1, 0]], dtype = float)

kernelVerySharp = np.array( [[ -1, -1, -1], [ -1, 9, -1], [ -1, -1, -1]], dtype = float)

# Morph kernels

largeMorph = cv2.getStructuringElement(cv2.MORPH_CROSS, (5, 5))

smallMorph = cv2.getStructuringElement(cv2.MORPH_CROSS, (3, 3))

# Capture video, resolution, FPS

video = cv2.VideoCapture(videofile)

(grabbed, I) = video.read()

height, width, _ = I.shape

fps = 24.0

# start timer for execution

startTime = time.time()

# 2)

# Inpainting mask

video.set(0,0)

(grabbed, I) = video.read()

I = cv2.cvtColor(I,cv2.COLOR_BGR2GRAY)

ret, mask = cv2.threshold(I, thresh = 10, maxval = 255, type = cv2.THRESH_BINARY_INV)

mask_inv = cv2.bitwise_not(mask)

video.set(0,0)

# 3)

# Remove watermark

print("STARTED REMOVING WATERMARK\n")

Image_data = []

while (video.isOpened()):

(grabbed, I) = video.read()

if grabbed is not False:

inpaint = cv2.inpaint(I,mask_inv,1,cv2.INPAINT_TELEA)

gray = cv2.cvtColor(inpaint, cv2.COLOR_BGR2GRAY)

Image_data.append(gray)

else:

break

print("FINISHED REMOVING WATERMARK\n\n")

# 4)

# Imagefile names and directory

directory = "./cleaned/"

imageExtension = ".png"

imagePrefix = "newframe"

if not os.path.exists(directory):

os.makedirs(directory)

# Clean video

makingVideo = True

fourcc = cv2.VideoWriter_fourcc(*'XVID')

newvideo = cv2.VideoWriter()

if makingVideo is True:

newVideo = cv2.VideoWriter(directory + 'ZorroClean.avi',fourcc, fps, (width,height))

# Frames

forceFrame = False

totalFrames = len(Image_data)

startFrame = 0

stopFrame = 0

frameNum = 0

if forceFrame is True:

startFrame = 105

stopFrame = 15

endFrame = stopFrame + startFrame

frameNum = 0

else:

startFrame = 0

stopFrame = totalFrames - 1

endFrame = totalFrames - 1

frameNum = 0

if makingVideo is True:

print("STARTED PROCESSING VIDEO\n")

else:

print("STARTED PROCESSING FRAMES\n")

if forceFrame is True:

print("GETTING FRAMES FROM " + str(startFrame) + " TO " + str(endFrame) + "\n")

else:

print("GETTING ALL FRAMES\n")

##################################################################################

# NOT FULLY MINE

#

# It shows percentage of progress as the program is executing

# Uses amount of frames in Image_data, compared to 'framepos'

#

# http://blog.montmere.com/2013/04/17/how-to-show-a-percent-complete-in-python/

##################################################################################

def percentage(fNum,fStop):

sys.stdout.flush()

# 5)

# processing algo

while (frameNum <= stopFrame):

percentage(frameNum,stopFrame)

currentFrame = startFrame + frameNum # current frame position

if currentFrame > endFrame:

break

frame = Image_data[currentFrame]

bilat = cv2.bilateralFilter(frame, 7, 75, 75)

sharpenBil = cv2.filter2D(bilat, ddepth = -1, kernel = kernelSharp)

dilated = cv2.dilate(sharpenBil,smallMorph)

eroded = cv2.erode(dilated,largeMorph)

dilatedFinal = cv2.dilate(eroded,smallMorph)

fastDen = cv2.fastNlMeansDenoising(dilatedFinal,None,3,3,9)

sharpNoise = cv2.filter2D(fastDen, ddepth = -1, kernel = kernelSharp)

sharpDil = cv2.dilate(sharpNoise,smallMorph)

sharpEro = cv2.erode(sharpDil,largeMorph)

sharpDilFinal = cv2.dilate(sharpEro,smallMorph)

gBlurred = cv2.GaussianBlur(sharpDilFinal,(3,3),0)

denoisedGray = cv2.fastNlMeansDenoising(gBlurred,None,8,7,21)

output = cv2.cvtColor(denoisedGray, cv2.COLOR_GRAY2BGR)

if makingVideo is True:

newVideo.write(output)

elif makingVideo is False:

imagename = directory + imagePrefix + str(currentFrame+1) + imageExtension

cv2.imwrite(imagename, output)

frameNum += 1

key = cv2.waitKey(1)

#if the 'q' key is pressed, quit:

if key == ord("q"):

break

video.release()

if makingVideo is True:

newVideo.release()

endTime = time.time()

print ("\n")

time = str("%.2f" % round(endTime - startTime,2))

print(time + " Seconds\n")

cv2.destroyAllWindows()