Cache Replacement Based on Similarity in Mobile Crowd Photographing

In recent years, with the development of smartphones and the spread of social network services (SNS) such as Twitter and Facebook, the mobile crowd photography (MCP), in which photos uploaded from smartphones are used for various services, has been widely used. For example, when a disaster occurs, rapid disaster relief is important to save human lives and reduce property loss. The key to efficiently and timely share and analyze the images is to determine the value or worth of the images based on their significance and redundancy, and only upload those valuable and unique images. While tens of millions of images are uploaded to the network, there is little need to deliver images that exactly match the user’s requirements, so it is important to be able to deliver images that are close to the desired images. Images are often delivered from cache servers such as CDNs and edge caches. Because of the large number of images on the network and the large number of similar images, there is a high degree of image redundancy. However, since the cache capacity is limited, a cache replacement method is needed to select images to be left in the cache when the cache capacity is exceeded. Typical cache replacement methods include LRU (Least Recently Used) and FIFO (First In First Out). In this paper, we propose a cache replacement method that preferentially deletes images with the largest similarity to other images in the cache. By grouping images based on their similarity in the cache, we aim to reduce the time required for similarity calculation and improve the cache hit rate by considering popularity. The performance of the proposed method is compared with that of LRU and FIFO, and the effectiveness of the proposed method is clarified.

I. Introduction

With the recent improvement of smartphone camera functions and the increasing use of social media (Twitter, Facebook, etc.), the use of mobile crowd photography (MCP), which means users upload photos on the network and share their lives with others, is increasing. A large number of similar images are uploaded on the networks, so the redundancy of image data on the network increases, and it increases the consumption of storage resources on the network and reduces the efficiency of retrieval and the quality of the user experience. Also, it is difficult to find images that exactly match the user’s requirements.