Appearance
Contents
Visual Preprocessing Report
The preprocessing of data is split into textual and visual preprocessing, with the visual preprocessing focusing on resizing the images to a smaller size for faster processing, normalizing their pixel values for later use by machine learning algorithms and storing them in unidimensional vectors.
Resizing, Normalization and Flattening
Uniform Sizing: All images in both the training and test sets were resized to a smaller fixed dimension of 224x224 pixels for less resource-consuming further processing.
Flattening: Each image was converted from a 3D array (height, width, channels) into a 1D vector since a flat feature vector is what most classical machine learning models will expect.
Pixel Value Normalization: Pixel values, originally in the range [0, 255], were scaled to [0, 1] by dividing by 255. This normalization improves model convergence and stability.
Saving as Numpy Arrays: The resulting 1D normalized vectors were saved as
.npyfiles for efficient loading and processing in subsequent modeling steps.
Grayscale Conversion
Images were converted to grayscale before flattening and normalization.
This reduces the number of channels from 3 (RGB) to 1, potentially simplifying the feature space and reducing computational requirements.
While reducing dimensionality, grayscaling may discard useful color information for certain tasks.