Deep Learning Algorithm for Video Saliency Object Detection using 3D DWT with Set Partition Integer Hierarchical Tree List
Suresh Babu D, Cyril Prasanna Raj, Professor, Dept of ECE, CIT, Bangalore. India · International Journal on Electrical Engineering and Informatics · 2021
In this work detection of salient objects in image and video sequences with higher accuracy, faster processing speed and reduced computation complexity is designed using deep learning algorithm.3 Dimensional (3D) Discrete Wavelet Transform is combined with Set Partition Integer Hierarchical Tree List (SPIHTL) encoding methods for identifying selfsimilarity coefficients of the salient object to be detected.Deep learning algorithm with seven layers is proposed in this invention that processes the quantized wavelet sub bands obtained after 2-level 3D inverse DWT reconstruction process detects the optimum number of features representing salient objects in the input data.An apparatus for evaluation of proposed method for salient object detection is designed in this invention that reconstructs the image from the features.Evaluation metrics are identified for measuring the performance of the proposed methods in this invention and are compared with traditional methods.The measure of PSNR and MSE for 120 different data sets of various dimensions and orientations demonstrates an improvement of 12%-18% in PSNR measurement as compared with existing methods.