A Robust Algorithm for Face Detection on Unconstrained Background Images using Neural Network

Abstract

Purpose: The main goal of proposed design is to use a neural network to create a robust solution for face detection on unconstrained backdrop images. In this regards the types of recognition involved in image processing discipline are pattern recognition, Face recognition, object recognition, speech recognition, and analyzing the data, etc. The basic concepts, structure and ideas involved in this recognition are discussed in this section. In the last two decades, object detection and recognition has become the most interesting and challenging research field for the researchers.

Design/Methodology/Approach: Developing a theoretical concept based on model building using the for Face recognition method and analysis. The need for Face recognition and detection arises due to demand for automatic surveillance systems, Human Computer Interface (HCI), etc. A fast processor is required in real-time implementation of implanted image processing applications. For these systems, an accurate and fast real-time implementation of face selecting is essential. Therefore, a novel approach for tracking and determining the face on a live and unmoving image is projected in this paper.

Findings/Result: Based on the developed model, the resultant efficiency of the selected algorithm can be predictable on the basis of Specificity, Accuracy & Sensitivity the result of evaluation is very high. Background, head movement, and neural network or incorrect training are some of the reasons. The efficiency of the selected algorithm can be predictable on the basis of Specificity, Accuracy & Sensitivity. The estimated values are 95.3%, 88.6% & 89.2% correspondingly.

Originality/Value: A novel approach for tracking and determining the face on a live and unmoving image is projected in this works. This projected process is applied and efficiently tested in the laboratory for different original images with glasses and without glasses taken by camera (specification of camera is Logitech USB, 30fps, 1600 x 1200 pixels). This process is built in a processor named open multimedia applications platform of 1 GHz and by Open CV libraries the proposed algorithm is developed. The rate of success in this selected algorithm is very high as the hardware is very much accurate and works at high speed which is best used for real time. The rate of success on selected algorithms is ninety-eight percent; the two percent unsuccessful rate in results is for some reason like background, head movement & due to neural network/ inappropriate training.