- Search the nearest neighbours of the images stored in data/images/
- on raw pixel data
- no image processing / feature scaling is done
- no dimension reduciton is done
- the Euclidean distance is computed on the raw pixel data
- after applying the Sobel operator to the raw pixel data
- on tiny images dataset
- on toy data
- describe results
- green pixel probelm
- after applying dimension reduction via PCA
- on raw pixel data
- after applying the Sobel operator to the raw pixel data
- on raw pixel data
- dimension reduction
- compute mean
- compute standard deviation
- compute covariance matrix
- compute pca
- plot S (from svd)
- image processing
- Sobel operator
- implement the Sobel operator
- include the Sobel operator as a filter in the knn search
- Sobel operator
- visualize the filtered images in the knn summary report
- query image
- query results
- ensure that the filtered image that is shown really has been used by knn
- solve the green pixel problem
- possible solutions
- user another weighting scheme, e.g. 0.3, 0.3, 0.3
- does not work for toy dataset
- no edges are found anymore (even on edges that are clearly visible)
- compute the Sobel operator on each color channel
- user another weighting scheme, e.g. 0.3, 0.3, 0.3
- possible solutions
- clarify: why are there no edges for green lines after applying sobel
- lines do disappear after image is converted into grayscale
- the formular is: 0.21r + 0.72g + 0.07b
- the green component is quite near to the noise -> green pixel problem
- lines do disappear after image is converted into grayscale
- show filtered query image in the summary report
- knn.py can write the filtered query image into a file
- integrated sobel from scipy
- improved performance of gen.py to generate toy dataset
- compute knn with sobel filter enabled
- created library imageprocessing.py for reading/writing images