Bangla Consonants.

Contexts in source publication

Context 1

... way. It is the key towards a better mechanism which can be time-efficient, effortless and productive. Though Bangla is a popular language, it does not have a proper OCR system compared to other languages such as English. Bangla as a language is complex and the writing structure is different from other languages. Bangla language has consonants ( Fig. 1), vowels ( Fig. 2), modified vowels ( Fig. 3) and around 170 compound characters (Fig. 4) [17]. Such complex writing structure needs better segmentation process for conversion into digital media, hence the applications for it is ...

Context 2

... from a book or document, the image may contain some portion outside of the text page. One of the challenges here is to identify the text of the image and crop the image so that the unwanted parts outside of the text can be eliminated. When binarized, these unwanted parts provide a chunk of black pixels which result in poor segmentation of lines. Fig. 10 shows unwanted chunk of black pixels marked in a red ...

Context 3

... solve the problem, the input image is cropped. For appropriate cropping of the image, we have performed page-layout analysis and have found out where the text is. Fig. 11 shows step by step procedures of how a text image is cropped which have been described ...

Context 4

... dewarping is associated to perspective correction. Geometric distortion of a captured image lines is a common real life scenario. The formation of curved lines due to view angle of camera or warped page leads to poor line segmentation, as most of the lines overlap with each other. As a result, multiple lines get segmented as a single line. Fig. 12 shows such a problematic scenario. The steps of our proposed image dewarp algorithm are as ...

Context 5

... is calculated depending on the shape of the span and an angle that it is distorted at. With the estimated parameter, coordinate transformation is done to make the lines parallel and horizontal. -Finally, we optimize the remapping of span to minimize the re-projection error using scipy.optimize.minimize which is a derivative-free optimizer. Fig. 13 shows an example of a geometrically distorted image before and after the use of image ...

Context 6

... are detected easily from an image horizontally. At first the image pixel values are calculated for each of the rows and are compared. Line segmentation is performed where the sum of the pixel value is close to zero (Fig. ...

Context 7

... we have faced another challenge while working with multiple font sizes in single page, where we fail to segment each of the lines properly. When multiple font sizes are present on the same image, line segmentation is performed for the bigger font size. As a result, all the lines are not segmented correctly. As shown in Fig. 15, the first two lines get segmented together due to different font ...

Context 8

... are segmented easily from segmented line images in a vertical manner. At first, the image pixel values are calculated for each of the columns of a segmented line image. Word segmentation is performed where the sum of the pixel value is close to zero (Fig. ...

Context 9

... separate each character, we need to detect matra line and then remove it. To properly detect the matra line, we have horizontally divided the word image into half. Matra line is detected where the sum of pixel value of rows are greater than 60% on the upper half of the image. Fig. 17 shows the region of Matra line for a ...

Context 10

... matra line, we get an open space between the characters as the characters are not connected with each other with the matra line. Characters then can be detected from an image vertically. At first the image pixel values are calculated for each of the columns. Character segmentation is performed where the sum of the pixel values is close to zero (Fig. 18). This is because, in a few cases, matra line is removed partially. Sometimes only a portion of matra line gets removed for which segmentation is done with sum of pixel value close to zero. Character segmentation is considered to be correct if a consonant or a vowel or a compound character is segmented alone or alongside with a modified ...

Context 11

... 18). This is because, in a few cases, matra line is removed partially. Sometimes only a portion of matra line gets removed for which segmentation is done with sum of pixel value close to zero. Character segmentation is considered to be correct if a consonant or a vowel or a compound character is segmented alone or alongside with a modified vowel. Fig. 19 shows examples of some correctly segmented ...

Context 12

... line, word and character accuracy of the multiple font step was not included from the image in fig. 20. The accuracy of this step was conducted from the image in fig. 15. The accuracy comparison of each of the methods clearly shows the importance of each step and how all these steps together is the key for a better segmentation ...

Context 13

... we straighten a curved line, a few words may stay tilted. In such cases, the matra line goes undetected, because the horizontal pixel sum criteria does not work here. This makes it harder for us to eliminate the matra line. Fig. 21 shows the region of matra line being detected of a slightly tilted word along with row-wise black pixel histogram. Fig. 22 shows the region of matra line which stays undetected and was not removed. As a result, we fail to segment such words into characters properly. Fig. 23 shows some examples where the character segmentation of the ...