Improving K-Nearest Neighbor Performance in Footwear Classification Using Leave One Out Cross Validation
Mayang Sari, Wikky Fawwaz Al Maki · 2023
Footwear is a tertiary need for humans in this world. The fashion world will continue to evolve, including the development of footwear fashion. In addition, based on the Criminal Investigation Agency of the Republic of Indonesia, footwear can be used as evidence of a crime, and can be used to identify a crime. Computer vision still has difficulty categorizing fashion, even though fashion plays a huge role in everyday life. Therefore, footwear image classification will be applied to help identify crimes and meet human needs in categorizing desired footwear and can be used to identify potential suspects, identify brands and models of shoes found at crime scenes. A series of methods are applied, starting from pre-processing, including image labelling and converting the image from RGB to grayscale, Histogram of Oriented Gradients (HOG) feature extraction, edge detection using the Canny Edge Detection algorithm, and to enhance the accuracy of KNN classification, the method applied is Leave One Out Cross Validation (LOOCV) the method works by splitting the data and save the accuracy in each iteration, which results in an increase in accuracy of 4%, from 94% using the KNN algorithm alone and with a combination of KNN and LOOCV resulting in the highest accuracy of 98%.