Short time fourier transform with coefficient optimization for detecting salient regions in stereoscopic 3D images: GSDU

• Published in: Multimedia tools and applications 2020-04, Vol.79 (13-14), p.8801-8824
• Main Authors: Rakesh Y, Sri Rama Krishna K
• Format: Article
• Language: English
• Subjects: Computer Communication Networks; Computer Science; Correlation coefficients; Data Structures and Information Theory; Divergence; Feature extraction; Fourier transforms; Gabor filters; Gaussian process; Identification methods; Image detection; Mathematical models; Model accuracy; Multimedia; Multimedia Information Systems; Optimization; Parameters; Salience; Special Purpose and Application-Based Systems; Stereoscopy; Texture
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-018-6686-x

In multimedia scenario, number of saliency detection designs has been portrayed for different intelligent applications regarding the accurate saliency detection like human visual system. More challenges exist regarding complexity in natural images and lesser scale prototypes in salient objects. In lots of the prevailing methods, the competency of identifying objects’ instances in the discovered salient regions is still not up to the mark. Hence it is planned to assist a new strategy by considering certain parameters of feature evaluation under optimization algorithms which diverts the new method of capturing the optimal parameters to acquire better outcomes. The given paper proposes a new saliency detection design that encompasses 2 phases like Feature extraction and depth saliency detection. In which Gaussian kernel model is processed for extracting the STFT features (Short-Time Fourier Transform), Texture features, and Depth features; and Gabor filter is used to get the depth saliency map. Here, the color space information is progressed under STFT model for extracting the STFT features. This is the major contribution, where all the STFT feature, Texture feature and Depth features are taken out to gain the depth saliency map. Additionally, this paper proffers a new optimization prototype that optimizes 2 coefficients namely feature difference amongst image patches from feature evaluation, and fine scale, by which more precise saliency detection outcomes can be attained through the proposed model. Subsequently, the proposed GSDU (Glowworm Swarm optimization with Dragonfly Update) contrasts its performance over other conventional methodologies concerning i) ROC (Receiver Operator Curve), ii) PCC (Pearson Correlation Coefficient), iii) KLD (Kullback Leibler Divergence) as well as iv) AUC (Area Under the Curve), and the efficiency of proposed model is proven grounded on higher accuracy rate.