KMK3483 Group Assignment by Syafisa Iskandar

Group Information

qsyafisaa — 2025-01-06 12:18:57 UTC

NUR SYAFISA QHAIRUNNISA (87166)

SITI FAZIRAH (86726)

NURUL HAYATI (85372)

ANIS NATASHA (83258)

MUHAMMAD SYAHZARULFADHIL (84758)

MOHAMMAD NADZMEE AZHAD

(84578)

Group Meeting

qsyafisaa — 2025-01-06 12:28:09 UTC

Group Discussion on GM

qsyafisaa — 2025-01-06 12:55:01 UTC

qsyafisaa — 2025-01-06 16:29:39 UTC

Thresholding

qsyafisaa — 2025-01-14 04:33:33 UTC

Displaying/Load Grayscale Images

qsyafisaa — 2025-01-14 04:34:25 UTC

Removing Noise by Median Filtering

qsyafisaa — 2025-01-14 04:34:57 UTC

Creating an Image Histogram

qsyafisaa — 2025-01-14 04:35:36 UTC

Adjusting Contrast Using Histogram Equalization

qsyafisaa — 2025-01-14 04:36:09 UTC

Compute the Magnitude of The Gradient Using Sobel

qsyafisaa — 2025-01-14 04:37:05 UTC

Compute the Magnitude of The Gradient Using Prewitt

qsyafisaa — 2025-01-14 04:37:56 UTC

Compute the Magnitude of The Gradient Using Gradient

qsyafisaa — 2025-01-14 04:38:33 UTC

Detecting Lines Using the Hough Transform

qsyafisaa — 2025-01-14 04:39:02 UTC

Pattern Finding in Image Using Cross-Correlation

qsyafisaa — 2025-01-14 04:39:39 UTC

Part C (ii) a. Use the CAN

2025-01-14 04:42:14 UTC

Video Image with Candy Line Detection

qsyafisaa — 2025-01-14 04:42:20 UTC

2025-01-14 04:42:37 UTC

2025-01-14 04:42:52 UTC

Output 1

qsyafisaa — 2025-01-14 04:43:37 UTC

Part C (i) Understand the CAN

2025-01-14 04:54:54 UTC

The number of training images used is determined by the command numImages = numel(imds.Files). The size of the input image is [256 256 3].

There are two types of images used for the training process: clean images and noisy images.

The inputs to the network are noisy images, while the desired responses of the network are the corresponding clean images.

The layers used in the multiscale CAN (Convolutional Autoencoder Network) from the Deep Learning Toolbox™ include the Input Layer, Convolutional Layers, Batch Normalization Layers, and Activation Layer.

The first layer of the CAN network is the Input Layer, the middle layer is a Convolutional Layer, and the last layer is an Activation Layer.

The training optimizer used to train the network is Adam. The default values of the network's hyperparameters are as follows:

Learning Rate: 0.001 (initial)
MiniBatchSize: 32
MaxEpochs: 10

qsyafisaa — 2025-01-14 04:57:54 UTC

What are the problem or limitation faced in this part of the project?

qsyafisaa — 2025-01-14 05:00:13 UTC

Some picture error to input in the axes picture

What are the problem or limitation faced in this part of the project?

qsyafisaa — 2025-01-14 05:00:21 UTC

Issue with Picture Placement on Axes
There was an error when trying to insert or overlay a picture onto the axes of a graph or chart. This could refer to difficulties aligning the picture, scaling it appropriately, or formatting it to fit correctly within the plot.

Group Comments

qsyafisaa — 2025-01-14 05:01:59 UTC

Our group think that this project aims to achieve robust, scalable, and accurate detection for applications such as industrial inspection, medical imaging, and object recognition.

Part C (ii) b. Use the CAN

2025-01-14 05:06:17 UTC

2025-01-14 05:07:38 UTC

2025-01-14 05:08:01 UTC

2025-01-14 05:08:28 UTC

2025-01-14 05:08:52 UTC

2025-01-14 05:09:07 UTC

Part C (ii) c. Use the CAN

2025-01-14 05:13:43 UTC

Noise images

2025-01-14 05:14:23 UTC

Denoised images obtained using multiscale CAN

PSNR Scores

2025-01-14 05:20:58 UTC

Group comments

2025-01-14 05:23:01 UTC

For part C, our group think that this project aims to demonstrate the power of deep learning in solving complex image processing challenges, with potential applications in fields such as healthcare, autonomous driving, and industrial automation.

What are the problems or limitations faced in this part of the project?

2025-01-14 05:23:11 UTC

Spend more time learning and understanding how to work on the CAN (Convolutional Autoencoder Network) part, as it can be a bit complicated.
Faced hardware issues because the program is large and needs a lot of processing power to run properly.