KNN Classification for CBIR with Color Moments, Connected Regions, Discrete Wavelet Transform
Kevin Song Mardi, Viny Christanti Mawardi, Novario Jaya Perdana · 2019
The difficulty of finding some specific images from a database creates a need for a system that applies the Content-Based Image Retrieval (CBIR) concept. Publicly, a CBIR system is a system that able to search and retrieve the desired images that have similarity in both visual characteristics and relevance terms. The proposed CBIR system is using three different feature extraction methods that is combined with the KNN Classification. The color feature will be extracted by Color Moments, the shape feature will be extracted by Connected Regions, and the texture feature will be extracted by Discrete Wavelet Transform. The images that will be used are animal images and separated into 10 classes. The KNN Classification will classify the query image first into a specific class before starting any retrieval process. The results from the system show that Color Moments has a precision of 45.8%, Connected Regions has a precision of 34.1%, and Discrete Wavelet Transform has a precision of 44.5%. Meanwhile, the combination of feature extraction methods without KNN Classification has a precision of 37.1%, while when combined with KNN Classification, the system has a precision of 68.7% with 100% classification accuracy. The experiment shows that the system needs KNN Classification to increase its usability and retrieval accuracy.