A Two Level Algorithm for Text Detection in Natural Scene Images

Abstract

In this paper we present a two-level method to detect text in natural scene images. In the first level, connected components (referred as CCs) are got from the images. Then candidate text lines are extracted and groups of connected components that align in horizontal or vertical direction are got. We think CCs in these groups have high probability are texts. To validate which CC is text, a SVM is trained to make an initial decision. The output of SVM is calibrated to posterior probability. Then we use the information of posterior probability of SVM and information of whether the connected component is in a group to divide the connected components into four classes: texts, non-texts, probable texts and undetermined CCs. In the second level, a conditional random field model is used to make final decision. Relationship between CCs is modeled by a network G(V, E), Vertices of the graph correspond to CCs. The determination in the first level will influence the second level's determination by giving different parameters of data term for the four classes of CCs. By this way, we not only use information of a single CC's feature, but also use the information of whether a CC is in a group to make final decision of whether the CC is text or nontext. Experiments show that the method is effective.

A Novel Text Location Algorithm in Complex Color Images

“…The edges between two vertices are called n-link. For the graph G, we define a similar standard energy to [14] in this paper, but there is a little different. The standard energy has two terms: data term and smoothed term:…”

“…These samples are found in internet. [14], we use restore error eR to classify character candidates into four kinds of texts: text, probable text, undetermined CCs, non-text. If character candidates fulfill heuristic rules in [8], we call them a group.…”

“…The V is a set of vertices, and E is a set of edges. The graph G is constructed using the method proposed in [14]. The vertices correspond to CCs, and the weights of edges in the graph correspond to standard energy.…”

“…The texts within natural images always contain large amounts of information. In order to analyze, recognize, comprehend the information better and apply them, such as the license plate retrieval, image/video retrieval and network filtering, and so on, many approaches for automatic detection and localization of text in images and videos have been proposed [1][2][3][4][5][6][7][8][9][10][11][12][13][14]. Existing approaches can be generally sorted into three categories: edge-based methods, texturebased methods, region-based methods.…”

“…Region-based methods [1,[8][9][10][11][12][13][14]] make use of the fact that a text line always contains similarly characteristics, e.g., the similar color. So, the clustering algorithm is used to quantify a color image into an image which only contains a few colors [8].…”