There is provided an image-based information extraction model, method, and apparatus, a device, and a storage medium, which relates to the field of artificial intelligence (AI) technologies, specifically to fields of deep learning, image processing, computer vision technologies, and is applicable to optical character recognition (OCR) and other scenarios. A specific implementation solution involves: acquiring a to-be-extracted first image and a category of to-be-extracted information; and inputting the first image and the category into a pre-trained information extraction model to perform information extraction on the first image to obtain text information corresponding to the category.

A method and apparatus for training a character detector based on weak supervision, a character detection system and a computer readable storage medium are provided, wherein the method includes: inputting coarse-grained annotation information of a to-be-processed object, wherein the coarse-grained annotation information including a whole bounding outline of a word, text bar or line of the to-be-processed objected; dividing the whole bounding outline of the coarse-grained annotation information, to obtain a coarse bounding box of a character of the to-be-processed object; obtaining a predicted bounding box of the character of the to-be-processed object through a neural network model from the coarse-grained annotation information; and determining a fine bounding box of the character of the to-be-processed object as character-based annotation of the to-be-processed object, according to the coarse bounding box and the predicted bounding box.

Complete removal of an underline which intersects a character may cause problems in a subsequent character recognition or conversion process, when parts of the character which coincided with the underline are also removed. To help reduce the problems, parts of underline may be removed from an image while parts of the character that coincide with the underline are maintained in the image. Areas where the character coincides with the underline are defined from a reduced version of the underline. When the underline is removed, the areas where the character coincide with the underline are maintained in a second image. The second image may then be subjected to a character recognition or conversion process with potentially fewer problems.

A text image processing method and a text image processing apparatus are provided. In some embodiments, a text image processing method includes: preprocessing a text image to obtain a binary image, where the binary image includes multiple connected regions; acquiring a convex hull corresponding to each of the connected regions with a convex hull algorithm; acquiring a character region circumscribing the convex hull; performing character segmentation on the acquired character region to obtain multiple character blocks; and merging the character blocks based on heights of the character blocks to obtain word blocks of the text image.

Computerized techniques for improved binarization and extraction of information from digital image data are disclosed in accordance with various embodiments. The inventive concepts include rendering a digital image using a plurality of binarization thresholds to generate a plurality of binarized digital images, wherein at least some of the binarized digital images are generated using one or more binarization thresholds that are determined based on a priori knowledge regarding an object depicted in the digital image; identifying one or more connected components within the plurality of binarized digital images; and identifying one or more text regions within the digital image based on some or all of the connected components. Systems and computer program products are also disclosed.

A local connectivity feature transform (LCFT) is applied to binary document images containing text characters, to generate transformed document images which are then input into a bi-directional Long Short Term Memory (LSTM) neural network to perform character/word recognition. The LCFT transformed image is a gray scale image where the pixel values encode local pixel connectivity information of corresponding pixels in the original binary image. The transform is one that provides a unique transform score for every possible shape represented as a 33 block. In one example, the transform is computed using a 33 weight matrix that combines bit coding with a zigzag pattern to assign weights to each element of the 33 block, and by summing up the weights for the non-zero elements of the 33 block shape.

Disclosed are methods for handwriting recognition. In some aspects, an image representing a page of a sample document is analyzed to identify a region having indications of handwriting. The region is analyzed to determine frequencies of a plurality of geometric features within the region. The frequencies may be compared to profiles or histograms of known language types, to determine if there are similarities between the frequencies in the sample document relative to those of the known language types. In some aspects, machine learning may be used to characterize the document as a particular language type based on the frequencies of the geometric features.

An image processing apparatus that generates an image for character recognition from a read image includes at least one memory that stores instructions, and at least one processor that executes the instructions to perform extracting of an area of handwritten character information and an area of printed character information from the read image, clipping of a partial image of the area of handwritten character information and a partial image of the area of printed character information out of the read image, and generating of the image for character recognition by combining the partial image of the area of handwritten character information and the partial image of the area of printed character information being associated with each other.

Methods and an apparatuses for recognizing a text, recognition devices and storage media are provided, which belong to the field of text detections. A method includes: extracting, by the recognition device, a feature map of a to-be-recognized image, then determining segmentation information of a text region of the to-be-recognized image based on a preset segmentation network and the feature map, and then determining boundary key points in the text region based on the segmentation information, and then converting a text in the text region into a text with a target arrangement sequence based on the boundary key points and then inputting the text obtained by conversion into a preset recognition model for recognition processing.

Systems and methods for extracting information from a sheet of material to facilitate organization of information from paper, and other physical writing surfaces are provided. An example system includes a sheet of material and a device for scanning the sheet with an optical sensor. The sheet of material includes an indication region. The indication region allows for indictors to be marked corresponding with at least one a corresponding subregion to be extracted. The sheet of material further includes at least one fiducial mark for identifying a boundary of the sheet. The device includes a processor operably coupled to the optical sensor for causing the optical sensor to scan the sheet and detect a boundary thereof using the fiducial marks and further identify a designated subregion of the sheet. Upon identification of the designated subregion, the processor is configured to extract information contained in the designated subregion for organization of information.