白丝美女被狂躁免费视频网站,500av导航大全精品,yw.193.cnc爆乳尤物未满,97se亚洲综合色区,аⅴ天堂中文在线网官网

System for character recognition in a digital image processing environment

專利號
US11176362B1
公開日期
2021-11-16
申請人
Bank of America Corporation(US NC Charlotte)
發(fā)明人
Madhusudhanan Krishnamoorthy; Nityashree Pannerselvam
IPC分類
G06K9/00; G06T5/10; G06T5/00; G06K9/26
技術(shù)領(lǐng)域
or,resolution,more,in,query,system,image,images,may,ibe
地域: NC NC Charlotte

摘要

Systems, computer program products, and methods are described herein for character recognition in a digital image processing environment. The present invention is configured to electronically retrieve one or more documents from a document repository, wherein the one or more documents are in an image format; initiate one or more image super resolution algorithms on the one or more documents; generate, based on at least the one or more image super resolution algorithms, one or more high-resolution images associated with each of the one or more documents; initiate one or more image bottleneck ensembles (IBE) algorithms on the one or more high-resolution images; extract, using the one or more IBE algorithms, one or more features associated with the one or more high resolution images; and store the one or more features extracted from the one or more high resolution images in a feature repository.

說明書

Next, as shown in block 206, the process flow includes generating, based on at least the one or more image super resolution algorithms, one or more high-resolution images associated with each of the one or more documents. In some embodiments, the image super resolution algorithms may be implemented using convolutional neural networks (CNN) due to its success in image classification and other applications in the computer vision fields, such as object detection, face recognition, and pedestrian detection. The efficient training implementation on modern powerful graphics processing unit (GPU) and the use of Rectified Linear Unit (ReLU) which makes convergence much faster while presenting good quality are some of the advantages of using CNN to implement the image super resolution algorithms. Accordingly, recovering a high-resolution image from a low-resolution image may be formulated using CNNs. The resulting image super resolution convolutional neural network (SRCNN) is configured to recover the high-resolution image of the document from its low-resolution counterpart in three steps, namely, patch extraction and representation, non-linear mapping, and reconstruction. In the patch extraction and representation step, the SRCNN extracts overlapping patches from the low resolution document. In response to extracting the patches, the SRCNN represents each patch as a high-dimensional vector. These vectors may comprise a set of feature maps, of which the number equals to the dimensionality of the vectors. Next, in the non-linear mapping step, the SRCNN non-linearly maps each high-dimensional vector onto another high-dimensional vector. Each mapped vector is conceptually the representation of a high-resolution patch. These vectors may comprise another set of feature maps. Next, in the reconstruction step, the high-resolution patch wise representations are aggregated to generate the final high-resolution image. In this way, the system may be configured to generate the one or more high-resolution images associated with each of the one or more documents.

權(quán)利要求

1
微信群二維碼
意見反饋