Uwimg

Uwimg is a C package for images processing that is callable in Python. It's an assignment of The Ancient Secrets of Computer vision CSE 455

Table Content

• Installation
• Basic Level Of Image Processing
  • Getting and setting pixels
  • Copy image
  • Grayscale image
  • Shifting the image colors
  • Clamping the image values
  • RGB and HSV
• Middle Level Of Image Processing
  • Image resizing
  • Image filtering with convolutions
    • Blurring
    • Blur and Resizing
    • Gaussian
    • Frequency and Reconstruction
  • Image Features
    • Sobel filters
    • Colorized Representation
• Panorama
  • Harris detector
  • Patch matching
  • Combine the images with a homography
• Optical Flow
• Neural Network

Installation

$git clone https://github.com/DiaaZiada/Uwimg
$cd Uwimg
$make

Original Image

Basic Level Of Image Processing

Getting and setting pixels

Getting pixel value

from uwimg import *
im = load_image("data/dog.jpg")
pixel = get_pixel(im, 0, 0, 0)
print(pixel)

=> 0.0470588244497776    

Setting pixel to a value

from uwimg import *
# remove all red colors from the image
im = load_image("data/dog.jpg")
for row in range(im.h):
    for col in range(im.w):
        set_pixel(im, col, row, 0, 0)
save_image(im, "output/no_red_dog")

Copy image

from uwimg import *
im = load_image("data/dog.jpg")
im2 = copy_image(im) 

Grayscale image

from uwimg import *
im = load_image("data/dog.jpg")
graybar = rgb_to_grayscale(im)
save_image(graybar, "output/gray_dog")

Shifting the image colors

from uwimg import *
im = load_image("data/dog.jpg")
shift_image(im, 0, .4)
shift_image(im, 1, .4)
shift_image(im, 2, .4)
save_image(im, "output/shifted_dog")

Clamping the image values

from uwimg import *
im = load_image("data/dog.jpg")
shift_image(im, 0, .4)
shift_image(im, 1, .4)
shift_image(im, 2, .4)
clamp_image(im)
save_image(im, "output/ligth_fixed_dog")

RGB and HSV

from uwimg import *
im = load_image("data/dog.jpg")
rgb_to_hsv(im)
shift_image(im, 1, .2)
clamp_image(im)
hsv_to_rgb(im)
save_image(im, "output/rgb_hsv_rgb_dog")

Middle Level Of Image Processing

Image resizing

Maximizing

Nearest Neighbor

from uwimg import *
im = load_image("data/dog.jpg")
a = nn_resize(im, im.w*4, im.h*4)
save_image(a, "output/4x_nn_dog")

Bilinear

from uwimg import *
im = load_image("data/dog.jpg")
a = bilinear_resize(im, im.w*4, im.h*4)
save_image(a, "output/4x_bl_dog")

Minimizing

Nearest Neighbor

from uwimg import *
im = load_image("data/dog.jpg")
a = nn_resize(im, im.w//7, im.h//7)
save_image(a, "output/7th-nn_dog")

Bilinear

from uwimg import *
im = load_image("data/dog.jpg")
a = bilinear_resize(im, im.w//7, im.h//7)
save_image(a, "output/7th-bl_dog")

Image filtering with convolutions

Blurring

from uwimg import *
im = load_image("data/dog.jpg")
f = make_box_filter(7)
blur = convolve_image(im, f, 1)
save_image(blur, "output/box7_dog")

Blur and Resizing

from uwimg import *
im = load_image("data/dog.jpg")
f = make_box_filter(7)
blur = convolve_image(im, f, 1)
thumb = nn_resize(blur, blur.w//7, blur.h//7)
save_image(thumb, "output/thumb_dog")

Gaussian

from uwimg import *
im = load_image("data/dog.jpg")
f = make_gaussian_filter(2)
blur = convolve_image(im, f, 1)
save_image(blur, "output/gauss2_dog")

Frequency and Reconstruction

from uwimg import *
im = load_image("data/dog.jpg")
f = make_gaussian_filter(2)
lfreq = convolve_image(im, f, 1)
hfreq = im - lfreq
reconstruct = lfreq + hfreq
save_image(lfreq, "output/low_frequency_dog.jpg")
save_image(hfreq, "output/high_frequency_dog.jpg")
save_image(reconstruct, "output/reconstruct_dog.jpg")

low frequency

high frequency

reconstruct

Image Features

Sobel filters

from uwimg import *
im = load_image("data/dog.jpg")
res = sobel_image(im)
mag = res[0]
feature_normalize(mag)
save_image(mag, "output/magnitude")

Colorized Representation

from uwimg import *
im = load_image("data/dog.jpg")
res = sobel_image(im)
mag = res[0]
feature_normalize(mag)
img = colorize_sobel(im)
save_image(img, "output/colored_magnitude_dog")

Panorama

Harris detector

from uwimg import *
im = load_image("data/Rainier1.png")
detect_and_draw_corners(im, 2, 50, 3)
save_image(im, "output/corners")

Patch matching

from uwimg import *
a = load_image("data/Rainier1.png")
b = load_image("data/Rainier2.png")
m = find_and_draw_matches(a, b, 2, 50, 3)
save_image(m, "output/matches")

Combine the images with a homography

from uwimg import *
im1 = load_image("data/Rainier1.png")
im2 = load_image("data/Rainier2.png")
pan = panorama_image(im1, im2, thresh=50)
save_image(pan, "output/easy_panorama")

Final panorama image

Optical Flow

from uwimg import *
a = load_image("data/dog_a.jpg")
b = load_image("data/dog_b.jpg")
flow = optical_flow_images(b, a, 15, 8)
draw_flow(a, flow, 8)
save_image(a, "output/optical-flow")

image a

image b

optical flow

Neural Network

Download data

wget https://pjreddie.com/media/files/mnist_train.tar.gz
wget https://pjreddie.com/media/files/mnist_test.tar.gz
tar xzf mnist_train.tar.gz
tar xzf mnist_test.tar.gz

Train script

from uwimg import *

def softmax_model(inputs, outputs):

	l = [make_layer(inputs, outputs, SOFTMAX)]

	return make_model(l)

def neural_net(inputs, outputs):

	print(inputs)

	l = [ make_layer(inputs, 32, LOGISTIC),

	make_layer(32, outputs, SOFTMAX)]

	return make_model(l)

print("loading data...")

train = load_classification_data("mnist.train", "mnist.labels", 1)

test = load_classification_data("mnist.test", "mnist.labels", 1)

print("done")

print

print("training model...")

batch = 128

iters = 1000

rate = .01

momentum = .9

decay = .0

m = softmax_model(train.X.cols, train.y.cols)

train_model(m, train, batch, iters, rate, momentum, decay)

print("done")

print()

print("evaluating model...")

print("training accuracy: %f", accuracy_model(m, train))

print("test accuracy: %f", accuracy_model(m, test))

loading data...
done

training model...
000000: Loss: 2.370572
000001: Loss: 2.302090
000002: Loss: 2.287877
...
...
...
000997: Loss: 0.347353
000998: Loss: 0.387637
000999: Loss: 0.330449
done

evaluating model...
training accuracy: %f 0.90235
test accuracy:     %f 0.9077