Handwritten Word Searching by Means of Speech Commands Using Deep Learning Techniques
| Autor: | Cesar G. Pachon-Suescun, Robinson Jimenez-Moreno, Javier Orlando Pinzón-Arenas |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
business.industry
Computer science Mechanical Engineering General Chemical Engineering Deep learning Speech recognition Image (mathematics) Task (computing) Handwriting recognition Modeling and Simulation Selection (linguistics) Preprocessor Artificial intelligence Electrical and Electronic Engineering business Implementation Word (computer architecture) |
| Zdroj: | International Review on Modelling and Simulations (IREMOS). 12:253 |
| ISSN: | 2533-1701 1974-9821 |
| DOI: | 10.15866/iremos.v12i4.17166 |
| Popis: | Word searching is a topic that has gained great interest due to the development of numerous techniques focused on this task. However, most of those implementations require a prior digitalization or preprocessing of the target text so, it cannot be done in real-time, or do not locate the word requested within the text or image, making the user search the word by their own. For that reason, in this paper, it has been proposed to develop an application focused on locating handwritten words in real-time, i.e. a webcam acquires the image continuously and the application searches for the word required by the user, indicating how many times the word is in the image and their respective location. The work has focused on detecting 10 words in the Spanish language. For this development, two recognition systems have been implemented. The first one is for speech recognition, in such a way that it is not required to enter the search by means of a keyboard, but that the selection is made using an audio input in charge of recognizing what word the user says; this is done by means of a convolutional neuronal network, whose accuracy has been 90% and is responsible for telling the application what the said word by the user. The second system is to detect and locate the words in the image acquired by the webcam, where a Faster R-CNN is used, validated with 98.9% accuracy in the words found. In order to verify the performance of the application, tests have been performed in real-time, showing the capacity it has, correctly identifying the word spoken by the user and locating with great precision each word found in the captured image. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|