Research On Color Image Reconstruction For Contact Image Sensor

Contact image sensor was first produced in the end of 1980's,and was first applied to scanner in 1998.It was applied to banking equipments such as cash registers in the beginning of the 21st century,and is still in the development stage.Recently,the research of the contact image sensor is focused on the improvement of the hardware,or the designing and optimization of the system which used the sensor.However,the research of the image of the sensor is still at the beginning.When used in banking equipment,there are some problems in the image of contact image sensor including resolution losing problem,divergence problem among the three channels,and the noise problem.This paper focuses on the problem of the images captured by contact image sensor which used in banking equipment,including resolution problem,divergence problem and noise problem,to improve the image's quality without cost increasing.Combined with the research of super-resolution and image demosaicing,we complete the following works:The structure and the imaging principle of contact image sensor are analyzed.And a model of the color image captured by contact image sensor is built,which is used to analyze the degeneration and the color pattern of the images.The problems to be solved are raised including resolution problem,divergence problem and noise problem.A group-based super-resolution algorithm with on-line dictionary learning is presented.The non-local self-similarity is used to divide the image blocks into groups.Then the on-line dictionary learning algorithm is used to combine the information of external database and the image itself so that a suitable dictionary will be trained for each group.At last,the image is reconstructed via sparse representation which will improve the quality of the result image.Experiments show that the result of this method is better than some state-of-the-art algorithms because of the sharper edges and better details.Our method will add information to the image so that the resolution of it will be rose.A color image reconstruct algorithm with guided filter is presented.The luminance channel which has been scaled up by any super-resolution algorithm was used as the guided image of the guided filter and then filters the chrominance channels which have been scaled up by interpolation.So that the chrominance channels will get the sharp edges and rich details from the luminance channel.Experiments show that our method can add information.And the quality of the color images will be further promoted.A reconstruction method for color contact image sensor image is presented,which is combined with super-resolution and image demosaicing algorithm.The inter-channel correlation is used as regularization term to make full use of the correlation between different channels,which offers information to the reconstruction to suppress the divergence problem and the aliasing effect.Experiments show that our method can weaken the aliasing effect to some extent,and will suppress the divergence problem.A fast reconstruction method for color contact image sensor image based on gray scale transformation is presented.This method focuses on the image used in banking equipment,which utilize the characteristic of the image to achieve a faster algorithm.The gray-scale transformation and normalized mutual information are used,so that the core operation is linear operation,which reduce the resource occupation and computation cost.This fast algorithm makes the realization in the banking equipment possible.Experiments show that this algorithm eliminates divergence effect,rises image resolution and achieves relatively superior result.We judge the reconstructed images in terms of both objective and human visual evaluations.Experiments on the simulated images and true images show that the algorithm we presented can solve the resolution problem,divergence problem and noise problem very well.In practical engineering application,the reconstruction results of our method help to raise the recognition rate of the character from 91%to 97%.We have applied the patent for our algorithm.