Abstract: Aspects of the present invention provide, in an optical/image character recognition (OICR) system comprising an OICR engine and a machine learning system, a method of training the machine learning system involving generation of degraded data for use in training the machine learning system. Other aspects of the present invention provide, in a similar OICR system, a method of restoring degraded end user data. Other aspects provide the OICR systems which function as described.

Abstract: Aspects of the present invention relate to a machine learning system that is trained to identify forms, performing a method that includes receiving a form as an input image; identifying a field in the input image; identifying boundaries of the field; identifying locations of characters in the field; creating a two-dimensional space containing special characters; replacing the special characters with the characters in the field; identifying one or more keywords in the field based on identification of words and/or location of words; and responsive to an indication that the identifying one or more keywords yielded an incorrect result, updating the machine learning system. In another aspect, the machine learning system is used to identify forms, and can identify whether a form requires registration and, if registration is required, performing the registration.

Abstract: Synthetic disease face image and disease facemask generation can provide training data for supervised learning of a variety of machine learning systems, including neural networks, which serve as detection models to detect disease or disorder affecting part or all of a person's face and/or cranium. Geometric transformations can be applied to facial images to generate the synthetic disease face images and disease facemasks.

Abstract: In different kinds of forms with incomplete lines, or with different color cells in lieu of lines, virtually completing or providing the lines enables formation of tables from which keywords and content in the forms may be identified. Where a form may have one or more such tables, as can be the case with forms with irregular formats, multiple tables may be identified, to facilitate identification of keywords and content in each such table. In embodiments, deep learning techniques may be applied. Cost analysis involving minimum distances between keywords and content may be performed, with the cost analysis also facilitating formation of a keyword dictionary and a content dictionary.

Abstract: A text correction method and apparatus can take advantage of a greatly reduced number of error-ground truth pairs to train a deep learning model. To generate these error-ground truth pairs, different characters in a ground truth word are replaced with a symbol, not appearing in any ground truth words, to generate error words which are paired with that ground truth word to provide error-ground truth word pairs. This process may be repeated for all ground truth words for which training is to be performed. In embodiments, pairs of characters in a ground truth word may be replaced with a symbol to generate the error words which are paired with that ground truth word to provide error-ground truth word pairs. Again, this process may be repeated for all ground truth words for which training is to be performed.

Abstract: A text correction engine meets different and changing end user requirements, with the ability to change a desired output by providing sufficient amounts of data, and by finetuning the appropriate text correction engine at the point of origin of the data. It is possible to retain confidentiality of data by retraining the base deep learning model at the base deep learning model's point of origin, to improve the base deep learning model's performance, making the base deep learning model more accurate for different contexts. Separate training of an end user model, leaving the base deep learning model intact, streamlines end user model training, and highlights desirable changes in the base deep learning model for further training or retraining.

Abstract: In some forms containing keywords and content, there may be nested levels of keywords, also referred to as a hierarchy. Content in the forms may be associated with one or more keywords in one or more of the nested levels, or in the hierarchy. Identifying keywords in adjacent cells in a table (with a nested keyword being either to the right of or below another keyword) enables distinguishing between keywords and content in filled forms, and enables correct association of content with respective keywords.

Abstract: In different kinds of forms with incomplete lines, or with different color cells in lieu of lines, virtually completing or providing the lines enables formation of tables from which keywords and content in the forms may be identified. Where a form may have one or more such tables, as can be the case with forms with irregular formats, multiple tables may be identified, to facilitate identification of keywords and content in each such table. In embodiments, deep learning techniques may be applied. Cost analysis involving minimum distances between keywords and content may be performed, with the cost analysis also facilitating formation of a keyword dictionary and a content dictionary.

Abstract: In some forms containing keywords and content, there may be nested levels of keywords, also referred to as a hierarchy. Content in the forms may be associated with one or more keywords in one or more of the nested levels, or in the hierarchy. Identifying keywords in adjacent cells in a table (with a nested keyword being either to the right of or below another keyword) enables distinguishing between keywords and content in filled forms, and enables correct association of content with respective keywords.

Abstract: A method and system to detect visual spoofing of a process of authenticating a person's identity employs computer vision techniques to define characteristics of different kinds of spoofing. Embodiments identify a foreground object within an image and by examining positions and/or orientations of that foreground object within the image, determine whether the presentation of the foreground object is an attempt to spoof the authentication process.

Abstract: Aspects of the present invention provide a new text location technique, which can be applied to general handwriting detection at a variety of levels, including characters, words, and sentences. The inventive technique is efficient in training deep learning systems to locate text. The technique works for different languages, for text in different orientations, and for overlapping text. In one aspect, the technique's ability to separate overlapping text also makes the technique useful in application to overlapping objects. Embodiments take advantage of a so-called skyline appearance that text tends to have. Recognizing a skyline appearance for text can facilitate the proper identification of bounding boxes for the text. Even in the case of overlapping text, discernment of a skyline appearance for words can help with the proper identification of bounding boxes for each of the overlapping text words/phrases, thereby facilitating the separation of the text for purposes of recognition.

Abstract: A method and system to detect visual spoofing of a process of authenticating a person's identity employs computer vision techniques to define characteristics of different kinds of spoofing. Embodiments identify a foreground object within an image and by examining positions and/or orientations of that foreground object within the image, determine whether the presentation of the foreground object is an attempt to spoof the authentication process.

Abstract: Aspects of the present invention provide a new text location technique, which can be applied to general handwriting detection at a variety of levels, including characters, words, and sentences. The inventive technique is efficient in training deep learning systems to locate text. The technique works for different languages, for text in different orientations, and for overlapping text. In one aspect, the technique's ability to separate overlapping text also makes the technique useful in application to overlapping objects. Embodiments take advantage of a so-called skyline appearance that text tends to have. Recognizing a skyline appearance for text can facilitate the proper identification of bounding boxes for the text. Even in the case of overlapping text, discernment of a skyline appearance for words can help with the proper identification of bounding boxes for each of the overlapping text words/phrases, thereby facilitating the separation of the text for purposes of recognition.

Abstract: In augmented reality (AR) and mixed reality (MR) representations of natural scenes that includes text on different kinds of surfaces, real-time text replacement facilitates user involvement with and appreciation of the natural scenes. Determination of surface curvature using a three-dimensional (3D) camera enables determination of consequent textual distortion and necessary compensation in order to read text accurately. Translation, transliteration, or other modification of text and replacement with that text in a natural scene enables a user to participate more fully in the scene.

Abstract: A 2D image is constructed from constituent 2D images that show different views of the same object. Construction is performed by taking image tiles, referred to as tonal triangles, from the constituent 2D images and combining them using 3D data for the object. The 3D data define a wireframe model comprising triangles, called contour triangles. Two tonal triangles are combined based on neighbor relationships between the contour triangles that correspond to those two tonal triangles. Additional tonal triangles may be combined as desired, until the 2D constructed image is of a size that shows the subject of interest. Compared to conventional processes for stitching and montaging, the process generates a 2D constructed image that is a more accurate presentation of the true area, shape, and/or size of the subject.

Abstract: In a color to grayscale image conversion particularly method suitable for processing color document images such as forms, the color image is analyzed to determined which of the red, green and blue channels are the most dominant, second most dominant, and least dominant channels, based on the amount of information contained in each channel. Then, coefficients are assigned to the three channels, where the coefficient for the most dominant channel is smaller than the coefficient for the second most dominant color channel, which is in turn smaller than the coefficient for the least dominant color channel. The grayscale pixel value is then calculated using a linear combination of the red, green and blue pixel values weighted by their assigned coefficients. In one example, the ratio of the coefficients for the least dominant, the second most dominant and the most dominant channels is 10:3:1.