Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

Assessment 3: Image Analysis

May 2023 (on campus students)

1. Apply the method of colour slicing to the image below to segment the blue  feathers of the duck. Explain the steps that you take, your choice of parameters and interpret the results. [2.5 points]

2. a) Apply the k-means clustering algorithm to segment the image below. Indicate the number of clusters that you use and justify your answer. [1.5 points]

b) Indicate whether the following statements are true or false. If they are false, explain why (no points will be given if no explanation is provided) [1 point].

i) The k-means clustering algorithm uses both information about the colour  of each pixel, as well as how pixels are connected.

ii) The watershed algorithm is based on the morphological operations of erosion and dilation.

3. The image below represents some blood vessels. Note that it exhibits white speckled noise on the background and black speckled noise within the blood  vessels. Which morphological operations can you apply to reduce that noise   (both in the background and within the blood vessel) and distort the image as little as possible? Apply them to the image, justify your approach and interpret your result. [2.5 points]

4. Construct a convolutional neural network to classify the MNIST image database with the following specifications: two convolutional layers, the first one consisting of 10 features maps, and a second one consisting of 15 feature   maps. The receptive field should be of size 5x5 for both convolutional layers. Use “ReLu” or “Ramp” as the activation function, and use max pooling layers of size 2x2 and stride 2. The fully connected layer at the end should have 3 layers, with 20, 15 and 10 neurons, respectively. Add a softmax layer at the end so that the output can be interpreted as the probability of belonging to a specific class (please note that the softmax layer at the end acts as the activation function,   and hence, we do not need to have another activation function just before it).  Train the network and calculate the confusion matrix for the test or validation  dataset. [2.5 points]