IMPLEMENTASI MACHINE LEARNING UNTUK MEMBUAT IMAGE SEARCH ENGINE BERDASARKAN ANALISIS SIMILARITAS [IMPLEMENTATION OF MACHINE LEARNING TO CREATE AN IMAGE SEARCH ENGINE BASED ON SIMILARITY ANALYSIS]

Abstract

The development of technology makes it easier to search for something on the internet. Based on data in 2022, there will be more than 1 billion Google users worldwide. This data proves the importance of search engines for internet users. However, in general, search engine searches use keywords. This research aimed to produce image search engines for searching. The image search engine utilizes a Convolutional Neural Network, namely VGG-16, which would compare the similarity of the input image to all images in the database by comparing the neurons in the image. Image search engine performance was measured based on the accuracy and sequence of image output. Accuracy was obtained with a weighting system from 5 search result images produced by the image search engine. Mobile phone applications were used to improve the quality of images used for searching. The application would capture images or take images through the gallery then upload them to Firebase and be used to search by image search engines. From the research conducted, it was found that changing the angle had the greatest impact on decreasing accuracy values when compared to 3 other factors: color, background and image quality. The decrease in accuracy due to the influence of angle, color, backround,, and image quality was 96.25%, 0%, 22.5%, and 68.75%, respectively. The images from cellphone application were proven to have higher accuracy (95.32% ) than the images captured with webcam (86.67%). The increase in accuracy emphasizes that the influence of image quality and the angle of the image used to search influences search results. The specifications of the cellphone camera are 64 Megapixels, while the webcam camera is 2 Megapixels