Smartphone camera document detection via Geodesic Object Proposals

Smartphones are enabling new ways of capture, hence arises the need for seamless and reliable acquisition of documents. We present an approach for identifying candidate document in a given image using Geodesic Object Proposals. The input images were downsampled to help to extract structures/features of interest and attenuate noise and increase runtime speed and accuracy. The results showed that it is promising to use Geodesic Object Proposals in document object detection task.

I. Introduction

In the past few years, the smartphone market expanded very quickly: according to the International Data Corporation [1], 2015 is likely to be the last year of double-digit smartphone growth. The 2015 calendar year finished with 1.44 billion smartphone shipments worldwide. As a consequence of users having available in any moment of their life a powerful device capable of capturing increasingly better pictures unleashed a new use: smartphones are progressively replacing desktop scanners, for personal or professional use. The Fig. 1 shows an example of a document detection by smartphone camera. Fig. 1.

Example document detection by camera-based smartphone