TextProposals: A text-specific selective search algorithm for word spotting in the wild

Abstract

Motivated by the success of powerful while expensive techniques to recognize words in a holistic way [1,2,3], object proposals techniques emerge as an alternative to the traditional text detectors. In this paper we introduce a novel object proposals method that is specifically designed for text. We rely on a similarity based region grouping algorithm that generates a hierarchy of word hypotheses. Over the nodes of this hierarchy it is possible to apply a holistic word recognition method in an efficient way.Our experiments demonstrate that the presented method is superior in its ability of producing good quality word proposals when compared with class-independent algorithms. We show impressive recall rates with a few thousand proposals in different standard benchmarks, including focused or incidental text datasets, and multi-language scenarios. Moreover, the combination of our object proposals with existing whole-word recognizers [2,3] shows competitive performance in end-to-end word spotting, and, in some benchmarks, outperforms previously published results. Concretely, in the challenging ICDAR2015 Incidental Text dataset, we overcome in more than 10 percent f-score the best-performing method in the last ICDAR Robust Reading Competition [4]. Source code of the complete end-to-end system is available at https://github.com/lluisgomez/TextProposals.

Words Matter: Scene Text for Image Classification and Retrieval

“…Our work is similar to the recent works [19], [15] in terms of providing word box proposals. [19] combines two generic object proposal outputs, namely Edge Boxes [67] and Aggregate Channel Feature Detector [8], as preliminary word box proposals.…”

“…state-of-the-art word box proposal methods [19], [15]. Third, we analyze the influence of ground-truth overlap threshold on word detection recall and word recognition accuracy.…”

“…Finally, the remaining word box proposals are processed using a convolutional neural network regressor to refine the coordinates of these word boxes. [15] performs an over-segmentation using maximally stable extremal region (MSER) algorithm with flexible parameters. Then, the segments are grouped together using distance metrics related to text (e.g.…”

“…In contrast, our word box generator is uniquely designed to detect text in images without any training involved. Moreover, [19], [15] in the end aim at high f-score word recognition whereas this work aims only at high recall. We experimentally verify high recall is more important than high f-score for the applications considered in this work.…”

“…We experimentally verify high recall is more important than high f-score for the applications considered in this work. Further, different from [19], [15] which address word recognition, the aim of this paper is to combine textual and visual cues for better fine-grained classification and logo retrieval.…”