Document Image Retrieval and Removal of Perspective Distortion Based on Voting for Cross-Ratios

We propose a method of document image retrieval using digital cameras. The proposed method takes as input parts or the whole of documents acquired as queries by a digital camera, and retrieve document images that include the queries. For this purpose, it is required to solve the problem of “perspective distortion” of images, as well as to establish a way for matching parts of document images flexibly. These are achieved based on the following characteristic points of the proposed method: (1) Indexing of document images using the projective invariant called the “cross-ratio”, (2) Voting with hashes whose indices are defined by the cross-ratios. We also propose a method of removal of perspective distortion using point correspondence obtained through the retrieval. This method is utilized for extracting annotation from document images. From experimental results using digital cameras with high resolution and a camera phone, we demonstrate the effectiveness of the proposed method.