Study On The Image Quality Assessment And Recognition Technology For Characters Pressed On Labels

Label is an important carrier of product information. The Characters pressed on the surface of label mark the product information, but there is no colour difference between these Characters and background, which makes the ways difference between the recognition of these characters and the general optical characters, so it difficult to use the existing research achievements of the traditional optical character recognition(OCR) to recognize them. With the development of the manufacturing product information, auto-recognition systems for these pressed characters become a necessary requirement for Product Information Management. Supported by Doctoral Fund of Ministry of Education of China and Shandong Province Natural Science Foundation, the research carried out in this thesis is as followings.The preprocessing methods for the label pressed characters'image using the wavelet packet transform are studied. The quality of the image preprocessing directly affects the success or failure of algorithms follow-up. Because the wavelet packet transform on image processing has many excellent characteristics, the methods of the label pressed characters'image denoising, smooth, edge detection, strengthen, and fusion using the wavelet packet transform are studied.The quality assessment method for the label pressed characters'image using the improved structural similarity index in the wavelet packet domain are studied. Image quality for the adequacy and accuracy of the obtained information plays a decisive role, so the building of an effective evaluation mechanism for the image quality is of great theoretical significance and application value. First, the subjective image quality assessment method are analyzed, and then the objective image quality assessment method are analyzed, and finally, the quality assessment method for the label pressed characters'image using the improved structural similarity index in the wavelet packet domain is focusedly studied.A novel feature extraction method based on characters'gray image using improved Singular Value Decomposition (SVD) in the wavelet packet domain is suggested. First, the images are decomposed by the wavelet packet transform (WPT), and then the singular values (SVs) feature and projection feature of approximation images obtained through the above process are combined to form a new eigenvector using weighting coefficient. The optimal image size and eigenvector dimension are selected by experimental method, and the comparisons of recognition experiments are carried out using Euclidean distance. Experiments confirm that the method could effectively improve the recognition rate of characters, and has much better robustness.The recognition methods for the characters pressed on the label are studied. The recognition methods for the characters pressed on the label using BP neural network, radial basis function neural network, and self-organizing neural networkare are studied. The recognition methods for the characters pressed on the label using support vector machines are studied.This work is supported by Doctoral Fund of Ministry of Education of China (20060422011) and Shandong Province Natural Science Foundation(Q2008G02).