Vocally specified text recognition in natural scenes for the blind and visually impaired

Abstract

Searching for items in our surrounding can be an instantaneous task for many. However, such task can consume much time and effort and cause frustration to the blind and the visually impaired. There is a great deal of information around us in textual form, such as signs and products’ labels that are not accessible by the blind. In most of the great efforts in assisting the blind and the visually impaired, a technology was introduced to identify hand-held objects by using a camera. It is not practical that the user must hold all items on a shelf until finding the desired item. The users need to search for what they want, not what the device or camera finds. This research is investigating how accurate could a developed application be in finding vocally specified text from natural scenes. To answer this research question, speech and text recognition modules can be combined to validate acquired data and measure the accuracy of locating vocally specified items. Enhancement techniques could be applied to individual modules to enhance the accuracy of combining both speech and recognition modules.