Next integrated result modelling for stopping the text field recognition process in a video using a result model with per-character alternatives

Abstract

In the field of document analysis and recognition using mobile devices for capturing, and the field of object recognition in a video stream, an important problem is determining the time when the capturing process should be stopped. Efficient stopping influences not only the total time spent for performing recognition and data entry, but the expected accuracy of the result as well. This paper is directed on extending the stopping method based on next integrated recognition result modelling, in order for it to be used within a string result recognition model with per-character alternatives. The stopping method and notes on its extension are described, and experimental evaluation is performed on an open dataset MIDV-500. The method was compares with previously published methods based on input observations clustering. The obtained results indicate that the stopping method based on the next integrated result modelling allows to achieve higher accuracy, even when compared with the best achievable configuration of the competing methods.

Fast Approximate Modelling of the Next Combination Result for Stopping the Text Recognition in a Video

“…For the validity of the stopping methods as it is described in [34] any metric function ρ between the text string recognition results can be used, provided that it satisfies the triangle inequality. For experiments in [34] and [35] a normalized version of the Generalized Levenshtein Distance (GLD) [37] was used. For calculating GLD a metric ρ C must be defined on the set of individual character classification results (on individual rows of the matrices).…”

“…The described stopping method was tested on text string recognition results without membership estimations [34], using ROVER [38] as a combination algorithm. Later it was tested with another per-frame recognition method and with an extended text string recognition result model [35] using a combination algorithm [30] based on a ROVER approach. In both experiments, it was shown that using such an approach for a given average number of processed frames the lower error level could be achieved and vice versa.…”

“…Optimizations proposed in subsections III-A and III-B consider a GLD as the distance function ρ. In this subsection we will consider a normalized version of the GLD [37], which was used for measuring a character-level recognition error rate in [34] and [35]. The normalized GLD always has a value in the range [0.0, 1.0] and satisfies triangle inequality.…”

“…In this section, we provide results of an experimental evaluation of the proposed optimization against the direct application of method evaluated in [35]. The source code and data necessary to reproduce the experiments are available at the following link: https://github.com/SmartEngines/stoppers modelling.…”

“…Section II provides an overview of the studied stopping method. The method was originally proposed in [34] and further evaluated in [35]. In section III the proposed approximations and optimizations are described.…”