Accuracy Enhancement of Portrait Segmentation by Ensembling Deep Learning Models

Abstract

Portrait segmentation is widely used as a preprocessing step in multiple applications. The accuracy of portrait segmentation models indicates its reliability. In recent times, portrait segmentation using deep learning models have achieved significant success in performance and accuracy. However, these portrait segmentation models are limited to a single model. In this paper, we propose ensemble approach using multiple portrait segmentation models to improve the segmentation accuracy. The result of experiment shows that the proposed ensemble approach produces better accuracy than individual models. Accuracy of single models and proposed ensemble approach were compared with Intersection over Union (IoU) metric and false prediction rate to evaluate the accuracy performance. The result shows reduced false negative rate and false discovery rate, this reduction in false prediction has enabled ensemble approach to produce segmented images with optimized error and improved result of segmentation in portrait area of human body than individual portrait segmentation models