PyVHR / evm.py
navagg's picture
Upload 45 files
39b4c8c
import argparse
import numpy as np
import cv2
import scipy.signal as signal
import scipy.fftpack as fftpack
def build_gaussian_pyramid(src, levels=3):
"""
Function: build_gaussian_pyramid
--------------------------------
Builds a gaussian pyramid
Args:
-----
src: the input image
levels: the number levels in the gaussian pyramid
Returns:
--------
A gaussian pyramid
"""
s = src.copy()
pyramid = [s]
print(s.shape)
for i in range(levels):
s = cv2.pyrDown(s)
pyramid.append(s)
print(s.shape)
return pyramid
def gaussian_video(video, levels=3):
"""
Function: gaussian_video
------------------------
generates a gaussian pyramid for each frame in a video
Args:
-----
video: the input video array
levels: the number of levels in the gaussian pyramid
Returns:
--------
the gaussian video
"""
n = video.shape[0]
for i in range(0, n):
pyr = build_gaussian_pyramid(video[i], levels=levels)
gaussian_frame=pyr[-1]
if i==0:
vid_data = np.zeros((n, *gaussian_frame.shape))
vid_data[i] = gaussian_frame
return vid_data
def temporal_ideal_filter(arr, low, high, fps, axis=0):
"""
Function: temporal_ideal_filter
-------------------------------
Applies a temporal ideal filter to a numpy array
Args:
-----
arr: a numpy array with shape (N, H, W, C)
N: number of frames
H: height
W: width
C: channels
low: the low frequency bound
high: the high frequency bound
fps: the video frame rate
axis: the axis of video, should always be 0
Returns:
--------
the array with the filter applied
"""
fft = fftpack.fft(arr, axis=axis)
frequencies = fftpack.fftfreq(arr.shape[0], d=1.0 / fps)
bound_low = (np.abs(frequencies - low)).argmin()
bound_high = (np.abs(frequencies - high)).argmin()
fft[:bound_low] = 0
fft[bound_high:-bound_high] = 0
fft[-bound_low:] = 0
iff=fftpack.ifft(fft, axis=axis)
return np.abs(iff)
def butter_bandpass_filter(data, lowcut, highcut, fs, order=5):
"""
Function: butter_bandpass_filter
--------------------------------
applies a buttersworth bandpass filter
Args:
-----
data: the input data
lowcut: the low cut value
highcut: the high cut value
fs: the frame rate in frames per second
order: the order for butter
Returns:
--------
the result of the buttersworth bandpass filter
"""
omega = 0.5 * fs
low = lowcut / omega
high = highcut / omega
b, a = signal.butter(order, [low, high], btype='band')
y = signal.lfilter(b, a, data, axis=0)
return y
def reconstruct_video_g(amp_video, original_video, levels=3):
"""
Function: reconstruct_video_g
-----------------------------
reconstructs a video from a gaussian pyramid and the original
Args:
-----
amp_video: the amplified gaussian video
original_video: the original video
levels: the levels in the gaussian video
Returns:
--------
the reconstructed video
"""
print(original_video.shape)
final_video = np.zeros(original_video.shape)
for i in range(0, amp_video.shape[0]):
img = amp_video[i]
print(img.shape)
for x in range(levels):
img = cv2.pyrUp(img)
print(img.shape)
img = img + original_video[i]
final_video[i] = img
return final_video