Snoopertext: A multiresolution system for text detection in complex visual scenes

Abstract

Text detection in natural images remains a very challenging task. For instance, in an urban context, the detection is very difficult due to large variations in terms of shape, size, color, orientation, and the image may be blurred or have irregular illumination, etc. In this paper, we describe a robust and accurate multiresolution approach to detect and classify text regions in such scenarios. Based on generation/validation paradigm, we first segment images to detect character regions with a multiresolution algorithm able to manage large character size variations. The segmented regions are then filtered out using shapebased classification, and neighboring characters are merged to generate text hypotheses. A validation step computes a region signature based on texture analysis to reject false positives. We evaluate our algorithm in two challenging databases, achieving very good results.

Robust text detection in natural images with edge-enhanced Maximally Stable Extremal Regions

“…As a primary test we use the well-known ICDAR text detection competition data set [26,15], which was also used as a benchmark for [16,27,28]. Further, we apply our algorithm to a document database, which we have created to test a document retrieval system based on text as well as low bit rate features in [29].…”

T-HOG: An effective gradient-based descriptor for single line text regions

“…We show that a support vector machine (SVM) classifier [2] using T-HOG descriptors can effectively solve the text/non-text classification problem. In particular, we show that the combination of a "permissive" text detector [3] with a T-HOG based post-filter outperforms state-of-the-art text detectors described in the literature [4]. We also show how the T-HOG could be used by itself in a top-down slidingwindow text detector, and as a component of an OCR system.…”

“…We processed each image collection with SnooperText [3], a state-of-the-art text detector algorithm, tuned for high recall and moderate precision. Through visual inspection, we separated the candidate regions returned by SnooperText into a set of text regions X i , and a set of non-text ('background') regions B i , for i = 1, 2, 3.…”

“…Minetto et al [3], which was developed within the iTowns urban documentation and navigation project [26]. SnooperText uses a multiscale adaptive segmentation to locate candidate characters, which are selected and grouped into words and lines by geometrical criteria.…”

Vehicle speed estimation by license plate detection and tracking

“…This detector follow the hypothesis generation and validation paradigm [30,42]. Namely, in the hypothesis generation phase we use the edge extraction, edge filtering, morphological dilation, and region grouping modules to provide coarse candidate regions based on the edge attribute that make up the license plate.…”