Wipro Limited

Prashanth Krishnapura Subbaraya

Raghavendra Hosabettu

This disclosure relates to method and system for optimizing artificial neural network (ANN) classification model and training data thereof for appropriate model behavior. The method may include extracting entities and domain specific entities from the training data for each of classes of the ANN classification mode, determining model parameters of the ANN classification model based on the training data, determining missing data with respect to the training data or the model parameters based on the entities and the domain specific entities for each the classes, iteratively analysing a relative advantage of a modified ANN classification model with a modified training data with respect to the ANN classification model with the training data, and determining an optimized ANN classification model and an optimized training data for appropriate model behavior based on the iterative analysis. The modified data may be generated based on the missing data.

Optimization of artificial neural network (ANN) classification model and training data for appropriate model behavior

The disclosure relates to method and system for improving performance of an artificial neural network (ANN) model. The method includes receiving the ANN model and input dataset. The ANN model includes neurons arranged in multiple layers and employing corresponding activation functions. The method further includes assigning a random activation threshold value to each of the corresponding activation functions, determining activated neurons in each layer for a majority of input data in the input dataset based on the random activation threshold value for each of the corresponding activation functions, identifying removable layers based on a number of activated neurons and a pre-defined threshold value, evaluating a relative loss of the ANN model upon removing each removable layer from the ANN model and for a random input data in the input dataset, and deriving a modified ANN model by removing one or more of the removable layers based on the evaluation.

Method and system for improving performance of an artificial neural network (ANN) model

This disclosure relates to method and system for improving performance of an artificial neural network (ANN). The method may include generating a weight matrix comprising weights of neural nodes in a given layer for each layer of the ANN, determining a marginal contribution value of a given neural node for each neural node in the given layer with respect to other neural nodes in the given layer, executing an elimination decision for each neural node in each layer based on the corresponding marginal contribution value, determining a distributed surplus value of a given remaining neural node in a given layer based on the marginal contribution values of a coalition of remaining neural nodes in the given layer for each remaining neural node in each layer, and updating the weight matrix based on the distributed surplus value of each remaining neural node in each layer.

Method and system for improving performance of an artificial neural network

This disclosure relates to method and system for improving classifications performed by artificial neural network (ANN) model. The method may include identifying, for a classification performed by the ANN model for an input, activated neurons in each neural layer of the ANN model; and analyzing the activated neurons in each neural layer with respect to Characteristic Feature Directive (CFDs) for corresponding neural layer and for a correct class of the input. The CFDs for each neural layer may be generated after a training phase of the ANN model and based on neurons in corresponding neural layer that may be activated for a training input of the correct class. The method may further include determining differentiating neurons in each neural layer that are not activated as per the CFDs for the correct class of the input based on the analysis; and providing missing features based on the differentiating neurons.

Method and system for improving classifications performed by an artificial neural network (ANN) model

This disclosure relates generally to image processing, and more particularly to method and system for reconstructing an image. In one embodiment, the method includes pre-processing an input image to generate character images corresponding to characters in the input image, determining a local character thickness threshold value for each character image, determining a global character thickness threshold value for the input image based on the local character thickness threshold values for the character images, and reconstructing each character image based on the local character thickness threshold value for each character image and the global character thickness threshold value to generate reconstructed character images. The local character thickness threshold value in a character image may be based on a set of character pixel values in a pre-determined number of segments in the character image. The method further includes re-constructing the input image based on the reconstructed character images.

System and method for reconstructing an image

This disclosure relates to method and system for improving performance of an artificial neural network (ANN). The method may include receiving a weight matrix comprising an original weight of each neural node in each layer of the ANN. For each unique combination of at least two neural nodes in each layer, the method may further include determining a relative advantage value for one of the at least two neural nodes in a given layer with respect to remaining of the at least two neural nodes in the given layer based on actual inputs and standard inputs to the at least two neural nodes, and determining a modified weight of each of the at least two neural nodes based on the relative advantage value. The method may further include executing an elimination decision for each neural node in each layer based on a corresponding final modified weight, and updating the weight matrix based on the final modified weight of each remaining neural node in each layer. The final modified weight of a given neural node in a given layer is the modified weight of the given neural node upon exhausting each unique combination of the at least two neural nodes in the given layer.

The present subject matter is related in general to the field of image processing, disclosing method and system for evaluating an image quality for Optical Character Recognition (OCR) Image evaluation system receives image comprising optical character data. The image evaluation system determines image parameter value for each of one or more image parameters of the image. The image parameter value for each of the one or more image parameters is determined for plurality of binary image segments identified in the image. The image evaluation system determines suitability value and impact value of the image, based on the image parameter value for each of the image parameters determined for the image. The image evaluation system determines quality score for the image, based on the suitability value and the impact value. The image is transmitted for processing before the OCR, upon determining the quality score to be above overall pre-defined threshold value.

Method and system for evaluating an image quality for optical character recognition (OCR)

A method and device for automatic data correction using context and semantic aware learning techniques is disclosed. The method includes extracting data within a document as machine readable text in a predefined format. The method further includes encoding each word of each line in the machine readable text to a multi-dimension word vector. The method includes generating a context word vector for each word in each line based on multi-dimension vectors associated with words succeeding and preceding the word in a line comprising the word. The method further includes decoding the context word vector associated with each word in each line to generate a corrected context word vector for each word. The method includes validating the corrected context word vector associated with each word in each line.

Method and device for automatic data correction using context and semantic aware learning techniques

This disclosure relates generally to document processing, and more particularly to method and system for detecting and extracting tabular data from a document. In one embodiment, the method may include generating a hierarchy of features, for a plurality of features of an image document derived from the document, based on relative spatial properties of the plurality of features. The method may further include segmenting the image document into a plurality of semantic segments based on the hierarchy of features, classifying each of the plurality of semantic segments into at least one of a plurality of tabular structures, and effecting at least one of a detection or an extraction of the tabular data from the image document based on the classification.

Method and system for detecting and extracting a tabular data from a document

A method and device for removing noise from an image is disclosed. The method includes creating a single dimensional vector for an image through a multi-layer neural network. The method further includes converting the single dimensional vector into a multi-dimensional matrix based on number of layers in the multi-layer neural network. The method includes generating a feature hierarchy based on the multi-dimensional matrix, such that the feature hierarchy comprises a plurality of levels, and each level in the plurality of levels comprises at least one feature associated with the image. The method further includes creating a plurality of segments for the image based on the feature hierarchy, such that each of the plurality of segments includes a set of features associated with the image. The method includes removing each segment comprising noise from the plurality of segments to generate a de-noised image.

Growing community of inventors

Bengaluru, India

Prashanth Krishnapura Subbaraya