Computer Vision by Nur Asmirah

MEMBERS

nrasmrh01 — 2024-01-07 16:06:16 UTC

Put your name and matric number <3

1)Input Image (Black & White)

2024-01-10 17:59:43 UTC

2)Output image 1 (Filling all holes)

2024-01-10 18:01:15 UTC

3)Output image 2 (Show all circular objects)

2024-01-10 18:02:25 UTC

4)Output image 3 (Show all circular objects with and w/o holes)

2024-01-10 18:03:29 UTC

5)Output image 4 (Show all rectangular objects)

2024-01-10 18:04:40 UTC

6)Output image 5 (Show all rectangular objects with and w/o holes)

2024-01-10 18:06:04 UTC

i. Understand the CAN - the multiscale context aggregation network.

2024-01-10 23:23:53 UTC

a. How many training images are used in the example? What is the size of the input image?

300, 000 images(20000 images x 15 epoch) with size of images 256-by-256 pixels.

b. There are two types of images used for the training process. What are these images?

pristineImages and bilatFilteredImages.

c. What are the inputs of the network? What are the desired responses of the network?

randomPatchExtractionDatastore

d. What are the layers used in the multiscale CAN from the Deep Learning Toolbox™?

adaptiveNormalizationMu — Scale layer that adjusts the strengths of the batch-normalization branch
adaptiveNormalizationLambda — Scale layer that adjusts the strengths of the identity branch

e. What is the first layer, middle layer, and the last layer of the CAN network of the example?

The first layer is image Input Layer that can operates on image patches. For the second layer is 2-D convolution layer. The last layer is a regression layer instead of a leaky ReLU layer.

f. What is the training algorithm used to train the network? What are the default values of the hyperparameter of the network?

Adam optimizer.
default values of 0.9 for "Momentum" and 0.0001 for "L2Regularization" (weight decay).

ii. b imnoise

lyvrance — 2024-01-15 09:33:45 UTC

Gaussian White Noise with code noisyImg = imnoise (img, 'gaussian');

imshow(noisyImg);

ii. b imbilatfilt

lyvrance — 2024-01-15 09:34:39 UTC

Bilateral Filtering with code : filteredImg = imbilatfilt(img);

imshow(filteredImg);

ii. a

lyvrance — 2024-01-15 09:38:40 UTC

The missing component are as follows:

Conv3-BN3-Mu3-add3/in1-Leaky3-Conv4
Conv4-BN4-Mu4-add4/in1-Leaky2-Conv5
Conv5-BN5-Mu5-add5/in1-Leaky2-Conv6
Conv6-BN6-Mu6-add6/in1-Leaky6-Conv7
Conv7-BN7-Mu7-add7/in1-Leaky2-Conv8
Conv8-BN8-Mu8-add8/in1-Leaky8-Conv9

i. Gray Scale Image

nrasmrh01 — 2024-01-15 10:40:46 UTC

ii. Remove noise by median filtering

nrasmrh01 — 2024-01-15 10:42:02 UTC

iii. Creating an Image Histogram

nrasmrh01 — 2024-01-15 10:42:50 UTC

iv. Threshold the image

nrasmrh01 — 2024-01-15 10:43:13 UTC

v. Adjusting intensity value using histogram equalization

nrasmrh01 — 2024-01-15 10:43:56 UTC

vi. Detecting edge using edge function

nrasmrh01 — 2024-01-15 10:45:16 UTC

vii. Detecting line using hough transform

nrasmrh01 — 2024-01-15 10:45:42 UTC

ii. c image 1 (original image)

nrasmrh01 — 2024-01-15 16:44:42 UTC

ii. c image 1 (denoised)

nrasmrh01 — 2024-01-15 16:45:29 UTC

denoiseImage = predict(net, noisyI);

ii. image 2 (original image)

nrasmrh01 — 2024-01-15 16:47:50 UTC

ii. image 2 (denoised)

nrasmrh01 — 2024-01-15 16:49:01 UTC

denoiseImage = predict(net, noisyI);

Design view

nrasmrh01 — 2024-01-15 16:58:57 UTC

Component

1 Axes

Image axes

8 Button

Normal image
Gray scale
Median filtering
Image histogram
Threshold
Histogram equalization
Edge function
Hough Transform

2 Edit field (text)

Name
Matric

Group whatsapp

nrasmrh01 — 2024-01-15 17:10:32 UTC