Vision and Machine Learning

1. In an object detection task using HOG, if the cell size is 8x8 pixels and the block size is 2x2 cells, and each cell histogram contains 10 bins, and the image size is 200x200 pixels, how many histograms will be generated in total?

• 2500

• 5260

• 5760

• 2580

In a real-time object detection system utilizing HOG features, to minimize the number of histograms generated for an image patch of size 96x96 pixels, what would be the optimal combination of cell size and block size?

• Cell size: 8x8 pixels, Block size: 3x3 cells

• Cell size: 12x12 pixels, Block size: 2x2 cells

• Cell size: 6x6 pixels, Block size: 2x2 cells

• Cell size: 10x10 pixels, Block size: 4x4 cells

Comparing two image patches, Patch A (120x120 pixels with a cell size of 10x10 pixels and a block size of 3x3 cells) and Patch B (100x100 pixels with a cell size of 5x5 pixels and a block size of 2x2 cells), which patch results in a higher number of histograms?

• Patch A

• Patch B

• Both have the same number of histograms

• Can't be determined

Analyzing the trade-offs between computational complexity and feature expressiveness in Histogram of Oriented Gradients (HOG), which adjustment would result in a higher number of histograms?

• Decreasing block size

• Decreasing the number of bins in the histogram

• Increasing image resolution

• Increasing cell size

Given a 5x5 window with the following intensity values, compute the Harris corner response function R for the central pixel of the window using the Harris corner detection algorithm. Assume a Gaussian filter with a standard deviation of 1 for smoothing, and a k-value of 0.04. (Select all correct statements.)

• The smoothed intensity values after applying the Gaussian filter are required for the computation.

• The Harris corner response function is calculated using the formula: R=λ₁λ₂−k(λ₁+λ₂)², where λ₁ and λ₂ are the eigenvalues of the structure tensor.

• The structure tensor is computed by convolving the image with the gradient of the image.

• The Harris corner response function is thresholded to identify corners in the image.

In the Harris corner detection algorithm, the k parameter is crucial for determining corners. If the k value is increased from 0.04 to 0.06, how would this affect the detection of corners in an image? (Select all correct effects.)

• Increasing k would make the algorithm more sensitive to corners.

• Increasing k would make the algorithm less sensitive to corners.

• Increasing k would increase the likelihood of false positives in corner detection.

• Increasing k would decrease the likelihood of false negatives in corner detection.

Compare the corner responses of two pixels in an image patch: Pixel A with intensity values [120,125,110,130,135] and Pixel B with intensity values [115,120,120,140,135], assuming the surrounding pixel intensities are the same for both. Which pixel is more likely to be identified as a corner according to the Harris corner detection algorithm? (Select all correct statements.)

• Pixel A is more likely to be identified as a corner because it has higher intensity values.

• Pixel B is more likely to be identified as a corner because it has a higher variation in intensity values.

• Pixel A is more likely to be identified as a corner because it has a higher Harris corner response value.

• Pixel B is more likely to be identified as a corner because it has a lower Harris corner response value.

The purpose of histogram equalization is to transform an input image such that its histogram resembles a ________ distribution, leading to improved visual quality.

Given a grayscale image with dimensions 200x200 pixels and a SIFT feature descriptor extraction process that utilizes a window size of 20x20 pixels and a 5x5 grid of subregions within the window, calculate the total number of SIFT descriptors that can be extracted from the image.

• 100

• 400

• 1125

• 2500

For a grayscale image with dimensions 256x256 pixels, determine the percentage of overlap between adjacent windows in a SIFT feature descriptor extraction process that uses a window size of 32x32 pixels, and step size 24x24 pixels.

• 6.25%

• 12.5%

• 25%

• 50%