Random base image representation for efficient blind vision

Abstract

Viola-Jones algorithm (Viola and Jones 2001; Int J Comput Vis 57(2):137–154, 2004) is a milestone in the development of face detection technology. It greatly improves the efficiency of face detection on the premise of ensuring the accuracy, which indicates that the research results in the field of computer vision have the ability to put into practical applications. In an application scenario, the client uploads photos to the client server which has trained Viola-Jones detector to detect objects. In the process, privacy leakage of both side is the problem to be solved. Clients want to protect their photos and the cloud server wants to protect its algorithm parameters. Blind Vision (Avidan and Butman 2006), introduced by Avidan & Butman, combined OT(Obvious Transfer) protocol with Viola-Jones face detector to construct a secure face detection protocol. However, the efficiency of Blind Vision is not ideal. It will take a couple of hours to scan a single image. In this paper, we proposed the Random Base Image (RBI) Representation based secure object detection method to speed the process. The original image is divided into several pictures which are sent to the cloud randomly. At the same time, random numbers and redundant fake classifiers are generated to protect privacy parameters of the cloud. Compared with the traditional Blind Vision (Avidan and Butman 2006) method, we did not use OT protocol and Secure Millionaire protocol to calculate feture response of classifiers and compare them with thresholds, but uses random numbers and redundant fake classifiers to protect the privacy of both sides. In addition, the integral-graph can be used to accelerate the calculation of random base images. Experiments show that our method can achieve the same detection accuracy as the original Viola-Jones detector and is much faster than the traditional Blind Vision (Avidan and Butman 2006) method.