Electronic Archive Khmelnitskiy National University ELARKHNU
Переглянути архів ІнформаціяMetadata
| Поле | Співвідношення |
| Title | Segmentation of text information in natural scene images |
| Names |
Kashtalian, A.
Каштальян, А.С. |
| Date Issued | 2017 (iso8601) |
| Abstract | Text detection and recognition is one of the difficult task in the computer vision, in particular in the case of images with a complicated background. The article is devoted to investigation of the task of text recognition in images with nonuniform background, in particular segmentation stage. The segmentation on lines, words and symbols are examined. The lines segmentation approach is based on determination of general intensity or color channels intensity and assumes definition of average intensity of the whole image, looking through every pixel line and definition its intensity, comparison line intensity with average intensity of image, finding of border between text and line spacing by intensity difference. The words segmentation approach is also based on determination of general intensity or intensity of color channels and consist of definition of text line general intensity, looking through every pixel column and definition its intensity, comparison with average text line intensity, finding the border between a word and space by intensity difference. The character segmentation based on finding maximally stable extremal regions (MSER) is suggested. The maximally stable external regions (MSER) feature detector works well for finding text regions because of the stable intensity profiles. This is a method for blob detection in images. The algorithm extracts from an image a number of covariant regions: a region is a stable connected component of some graylevel sets of the image. The MSER extraction implements the following steps: sweep threshold of intensity from black to white, perform a simple luminance thresholding of the image; extract connected components (extremal regions); find a threshold when an extremal region is “maximally stable”, i.e. local minimum of the relative growth of its square. The MSER detector marks out most of the text, it also detects many other stable regions in the images that are not text. These candidates are then filtered using regions geometric properties and stroke width information to exclude nontext objects. The segmentation process allows to extract characters for father classification and to create a dataset necessary for classifier training. The coding has implemented in python and qualitative analysis is performed. |
| Genre | Стаття |
| Topic | optical character recognition |
| Identifier | Kashalian, A. Segmentation of text information in natural scene images [Текст] / A. Kashalian // Вісник Хмельницького національного університету. Технічні науки. – 2017. – № 6. – С. 106-110. |