In today's world, there have been lots of unique optical character recognition systems. One drawback of these systems is that they cannot work effectively on natural scene images where the text is not only subject to different orientations, lightning, and background but can be of multiple scripts as well. The paper, proposes a state of the art algorithm to detect texts of different dialects and orientations in an image. The whole text detection pipeline is divided into two parts. First, extraction of probable text regions in an image is performed based on a combination of statistical filters, which results in a high recall. These regions are then fed to an Artificial Neural Networks (ANN) based classifier which classifies whether the proposed regions are text or non-text, which increases the overall precision. The validity of the algorithm is verified on the most challenging bilingual text detection dataset MSRA-TD500 and a promising F1 score of 0.67 is reported. © 2019-IOS Press and the authors. All rights reserved.