Research On Two-Dimensional Barcode Image Super-Resolution Restoration Via Sparse Representation

With the development of Information technology, obtaining useful information throughrecognizing two-dimensional barcode has become very common. The barcode has a large amountof information and a high density, so a high-resolution image is needed to insure its recognition.For example, some applications need the user to shoot a barcode image and then upload it to theserver for recognition. However, since these users lack sufficient guidance, it tends to make theuploaded image be a low-resolution one, which results in an unsuccessful recognition. Thus it canbe seen that it is important to conduct a super-resolution restoration process before barcoderecognition. Therefore, this paper has done a deep research on two-dimensional barcode imagesuper-resolution restoration via sparse representation. The main work of this paper is divided into:1) It discusses two major super-resolution restoration methods. This paper explains the super-resolution method based on reconstruction and based on learning. And it is found that the barcodeimage has excellent structures which make it convenient to establish the relationship betweenhigh-resolution image patches and low-resolution ones. Therefore, we conjectures the methodbased on learning is more suitable for our reconstruction.2) It explores the sparse representation model based on over-completed dictionary. First, thispaper analyzes its whole modeling process. Then, it gives a deep analysis for this model fromthree aspects: solution of the sparse model, measurement of sparsity and uniqueness of sparsecoefficients. At last, this paper illustrates the effectiveness of the model in image super-resolutionrestoration area.3) To build a sparse association between high-resolution image patches and low-resolutionones, it proposes a method of extracting feature, which is suitable for two-dimensional barcodeimage super-resolution reconstruction. The edge of barcode contains a large amount of high-frequency information, so this paper uses the Kirsch operator to detect the edge gradient as areconstruction feature. Gradient direction well reflects the texture information of barcode, so thispaper uses the advanced version of edge texture histogram method to extract gradient direction asanother reconstruction feature. At last, combining the features above with the vertical and thehorizontal second-order gradient, we get the final feature extractor for low-resolution patches.4) Using the extracting method above, it proposes a super-resolution restoration algorithm of two-dimensional barcode image. First, we use the Shock filter to enhance the edge of low-resolution training images so that features extracted are more accurate. Then, we use samplesbased on content to train the sparse dictionary, so that the dictionary has a stronger sparserepresentation ability. At last, we conduct a global constraint on the reconstructed image, so thatthe final estimated image is closer to the original one. Experimental results show that thisalgorithm can effectively achieve the super-resolution reconstruction of barcode image.