HU‐PageScan: a fully convolutional neural network for document page crop

Abstract

November The offer of online, automated, and impersonal services demand users to upload scanned copies of their documents to the organisations. As a consequence of this decentralisation, the documents present more challenges to the already complex process of image processing and information extraction. To address this problem, the authors presented an optimised fully convolutional neural network model for document segmentation that works on mobile devices to detect the region of the document in the captured image. They performed experiments in three representative datasets comparing the proposed method with the Geodesic object Proposals, U-net, Mask R-CNN, and OctHU-PageScan algorithms. They also compared the proposed model with all competitors of the ICDAR2015 Competition on smartphone document capture. Furthermore, they performed a qualitative and comparative analysis with the CamScanner software, a popular app for Android and iOS smartphones used for more than 100 million users in over 200 countries. The proposed approach achieved a significant performance compared with the current state-of-the-art methods, providing a powerful approach for document segmentation in photos and scanned images.

MIDV-2020: a comprehensive benchmark dataset for identity document analysis

“…The task is to clas-sify each pixel to either document or a background. The most natural metrics to evaluate such a task is to compute a pixel-wise Jaccard score [15].…”

“…To evaluate the semantic segmentation method, we used the published methods HU-PageScan [15] and HoughEncoder [16], both based on a U-net modification. HU-PageScan has an openly available source code and a pre-trained model, which was used for this evaluation.…”

“…Since its publication, MIDV-500 dataset and its extension MIDV-2019 were used to evaluate the methods of identity document images classification [12 -14]; identity document location [11,15], including the methods based on semantic segmentation [16]; detecting of faces on images of identity documents [17]; and methods related to text fields recognition, including single text line recognition [18], per-frame recognition results combination [19,20] and making a stopping decision in a video stream [21,22]. The dataset was also used to evaluate the methods of choosing a single best frame in the identity document video capture [23] and assessing the quality of the frame for its processing by an identity analysis system [24], detection and masking of sensitive and private information [25] and general ID verification [26].…”

Advanced Hough-based method for on-device document localization

“…The systems are listed in descending order by the average accuracy for all four SmartDoc subsets. The results clearly show that based on the overall accuracy, our system is the second best, surpassed only by [19] proposed by das Neves et al This method is based on a Unet like neural network. On the Fig.…”

“…There are millions of parameters in such networks, so researchers are concentrating on the reduction of the parameters. Ricardo et al in [19] modified the U-net network architecture, reducing the number of parameters by more than 70 %. In [20], the number of parameters was reduced by a factor of 100 via the Fast Hough Transform [21] and its inverse.…”