رسائل ماجيستير
English
2016
PRINTED NUMBER RECOGNITION USING ANDROID SYSTEM
قاسم محمد جاسم, وسام سمير بهيه
كلية تكنولوجيا المعلومات-جامعة بابل · العراق
To perform number recognition, the proposed work consists of three stages: preprocessing stage, segmentation stage, and recognition stage. In the preprocessing stage, a color image is converted into a grayscale image, and the edges of objects are detected by applying the Sobel operator, then noise pixels are removed using opening morphology.
The next stage is segmentation, which aims to isolate the candidate objects to be numbers in separated candidate objects. Here, connected component labeling is employed for segmenting the whole image into individual objects. Then, we perform projection for further segmentation after binarization, which depends on local threshold and automatic exchange of background and foreground color according to the writing line color.
Recognition is the last stage in this work, and it consists of three steps. At first, normalization is performed, and zoning feature analysis is applied as a filter to reduce the candidate objects. Then, features are extracted; two types of features will be extracted. Inner-outer contour tracing labeling technique is used for extracting inner and outer tracks. The inner will be the number of holes, and the outer will be the directions of the digit tracks. These features are used to recognize the object as a digit and classify it according to its label. At the end, structure features, which are selected from outer contour tracing, are used to ensure the digit is not an alphabet letter.
In this thesis, real-time number recognition has been implemented in an Android smartphone environment, specifically on the Samsung Galaxy Note 3. Also, the proposed work was performed on a laptop device. The result of recognizing 14 color images is 85% accuracy. These images have multiple backgrounds and foregrounds, different fonts, different sizes, and multiple natural scenes.