A faster color-based clustering method for summarizing photos in smartphone

Abstract

Many people use smartphones which feature cameras and large storage capacities. They have comfortable features that allow the recording of daily life and encourage them to take hundreds of photos. However, their limited screen sizes cause browsing problems with large photo set. In addition, organizing photos by manual is a hard and tediuous work. Therefore, a way to classify photos to events and a better visualization method are needed. Much research has automatically classified photos to easily present and manage collections. However, most algorithm were designed for computers, not for smartphones which have limited computing power. This paper describes a fast and lightweight spatial clustering method to summarize and visualize smartphone photos by event. We defined a nearly identical photos which are taken by people in order to get better quality photos. These photos do not need to visualize duplicately since they show same event. We calculated each photo's similarity and focused on eliminating inefficient steps to measure it. The CIELAB color space was selected to accurately measure color differences and quantize colors. We extracted nearly identical photos and cut events by optimal matching with each photo's color histogram. Our method changed the color quantization technique and accelerated clustering speed by >;20 while retaining precision and recall.