Saurav Mukherjee, Shanmukh Wankhede, Bhavesh Mishra, Prof. K. S. Chandwani
Student, Dept. of Computer Technology, KDK College of Engineering, Nagpur, India, saurav.mukherjee970@gmail.com
Student, Dept. of Computer Technology, KDK College of Engineering, Nagpur, India, Bhavesh.nm99@gmail.com
Student, Dept. of Computer Technology, KDK College of Engineering, Nagpur, India, shaanwankhede@gmail.com
Professor, Dept. of Computer Technology, KDK College of Engineering, Nagpur, India, chandwani1@rediffmail.com
Abstract:
Despite having plenty of resources of writing technologies many people from every profession depend on taking notes in a form of paper and pen. And it becomes difficult to maintain and access the notes in a systematic manner. If we have any technology that would convert the handwritten notes in digital format. Then the efficiency of storing handwritten notes will get increased. In this project, the technology will convert every handwritten character in a digital form, it will be helpful to store the data in the long run. This process involves two main approaches first, it processes the image to classify the handwritten characters and captions the handwritten characters. Second, it uses the Convolution Neural network (CNN) to train the model that will process every pixel of the captioned image and train the neural network i.e. Learning algorithm to predict the handwritten character and thus convert in digital form
Index Terms- Convolution neural network (CNN), Deep Learning, Handwritten character recognition, Image processing.
I. INTRODUCTION
Despite having plenty of resources of writing technologies many people from every profession depend on taking notes in a form of paper and pen. And it becomes difficult to maintain and access the notes in a systematic manner. If we have any technology that would convert the handwritten notes in digital format. Then the efficiency of storing handwritten notes will get increased. In this project, the technology will convert every handwritten character in a digital form, it will be helpful to store the data in the long run. In this process of handwritten character recognition involves Handwriting analysis and Handwritten characters characterization and captioning is implemented using image processing and image data classification. the Convolution Neural network (CNN) to train the model that will process every pixel of the captioned image and train the neural network i.e Learning algorithm to predict the handwritten character and thus convert in digital form. CNN will have multiple layers in the neural network that will process the data and give meaningful insights from the input data. In this approach, the Machine Learning (ML) algorithm will automatically improve its performance through the experienced. And based on the experienced the model will predict the handwritten characters with higher efficiency.
II. WORKING MECHANISM
Collect the dataset, organize and manage the annotation to caption(label) the data. In this process, we will create a Dataset that will contain data of the different handwritten text from different people and it will form collectively form a large dataset of handwritten characters datasets. In this process after acquiring Datasets of Handwritten characters, the model will annotate i.e. label or caption the handwritten text. After acquiring the datasets, it involves training those datasets. Datasets are required to combine the datasets. Then filter those datasets to remove unnecessary noises present in the training data. After removing noise from the data and pre-processing the data we can fetch those data to the Neural Network. And after fetching the Dataset and Captioning the text it will fetch the data to the Convolution Neural Network (CNN). Then the CNN will train the label data and help the API to predict the handwritten characters.
Train the customs neural network and track metrics in this process we apply a learning algorithm to train the neural network to train the model with large data. In this process efficiency of the model is said to learn from experience E concerning some Task measure T and the performance of measure P. The performance of the neural network will increase leading to higher prediction rates of the handwritten characters been classified by the neural network. Choose the best models and combine them into production pipelines in the production pipelines we are first applying the model to caption or annotate the image so that we can acquire the Data from the image. Then we will apply the second model that will classify and analyze the annotated document and In the final model, the model will convert the annotated handwritten text into digital form. Integration to our custom production environment. In this process, we will deploy this whole model in a single integrated API that will take input from the user to convert the data into a raw format to apply a neural network to predict the handwritten characters and it will simply convert those data in digital form.
III. Flowchart
IV. MODULE SCREENSHOTS
This is the module where the data is going to be available in the tabular format. The database is hosted using the MySQL sever and owned by Oracle Inc. The safety of the database is vitally important.
V. RELATED WORK
An early notable try within the space of character recognition analysis is by Grimsdale in 1959. The origin of an excellent deal of analysis add the first sixties was supported AN approach called analysis-by-synthesis methodology steered by Eden in 1968. the nice importance of Eden’s work was that he formally well-tried that every one written characters area unit shaped by a finite variety of schematic options, some extent that was implicitly enclosed in previous works. This notion was later employed in all strategies in syntactical (structural) approaches of character recognition. K. Gaurav, Bhatia P. K. [5] Et al, this paper deals with the various pre-processing techniques concerned within the character recognition with completely different quite pictures ranges from a simple written kind primarily based documents and documents containing coloured and sophisticated background and varied intensities. In this, completely different pre-processing techniques like skew detection and correction, image sweetening techniques of distinction stretching, binarization, noise removal techniques, normalization and segmentation, morphological process techniques square measure mentioned we tend to can’t fully method the image. However, even once applying all the aforementioned techniques might not attainable to attain the complete accuracy during a pre-processing system.
VI. CONCLUSION
The conclusion of this investigation is to execute a component extraction A methodology which can be utilized in any character acknowledgment Problem. Here, we have demonstrated that if a CNN is prepared for a sufficiently enormous class issue, it very well may be utilized for extraction of highlights of some other character set and the subsequent framework Is as yet fit for giving high acknowledgment correctness’s.
VI. FUTURE SCOPE
- Translation of handwritten stuff in other languages
- Solve handwritten mathematical equation or question just by placing camera over the question
- Handwritten characters recognition with natural language processing can take this process to next level of advancements
ACKNOWLEDGEMENT
We would like to show our gratitude to the professors for sharing their pearls of wisdom with us during the course of this research, and we thank them for insights and for their comments that greatly improved the report. We are also immensely grateful to our department for providing all the necessary equipment and facilities.
REFERENCES
- Puigcerver, Joan. “Are Multidimensional Recurrent Layers Really Necessary for Handwritten Text Recognition?.” Document Analysis and Recognition (ICDAR), 2017 14th IAPR International Conference on. Vol. 1. IEEE, 2017.
- Huang, B.; Zhang, Y. and Kechadi, M.; Preprocessing Techniques for Online Handwriting Recognition. Intelligent Text Categorization and Clustering, Springer Berlin Heidelberg,
- 2009, Vol. 164, “Studies in Computational Intelligence” pp. 25–45.
- D. C. Ciresan, U. Meier, J. Schmidhuber. Multi-column Deep Neural Networks for Image Classification. IEEE Conf. on Computer Vision and Pattern Recognition CVPR 2012.
- 2012 Kurzweil AI Interview Archived 31 August 2018 at the Wayback Machine with Jürgen Schmidhuber on the eight competitions won by his Deep Learning team 2009-2012
