Abstract
Quality assessment of stitched images is an important element of many virtual reality and remote sensing applications where the panoramic images may be used as a background as well as for navigation purposes. The quality of stitched images may be decreased by several factors, including geometric distortions, ghosting, blurring, and color distortions. Nevertheless, the specificity of such distortions is different than those typical for general-purpose image quality assessment. Therefore, the necessity of the development of new objective image quality metrics for such type of emerging applications becomes obvious. The method proposed in the paper is based on the combination of features used in some recently proposed metrics with the results of the local and global image entropy analysis. The results obtained applying the proposed combined metric have been verified using the ISIQA database, containing 264 stitched images of 26 scenes together with the respective subjective Mean Opinion Scores, leading to a significant increase of its correlation with subjective evaluation results.
Highlights
One of the methods recently proposed for quality assessment of stitched panoramic images, verified using the ISIQA database, is the Stitched Image Quality Evaluator (SIQE) proposed by the authors of this dataset [3]
This method utilized a comparison of 36 features calculated for the constituent and stitched images, namely the eigenvalues determined for the covariance matrix where the covariances are computed for each patch for the pair of wavelet coefficients for a bivariate distribution determined from the Gaussian Mixture Model (GMM) and shape parameters of the Generalized Gaussian
The initial experiments, conducted using 264 stitched images obtained for 26 scenes that are included in the ISIQA dataset as well as some additional stitched images generated using the freeware Hugin software with various parameters, have demonstrated the potential improvements of existing metrics caused by their diversity
Summary
Panoramic images, constructed as a result of image stitching operation conducted for a series of constituent images with partially overlapping regions, may suffer from various distortions, including blur, ghosting artifacts, and quite well visible geometric and color distortions. During several recent years a great progress has been made in generalpurpose image quality assessment (IQA), the direct application of those methods proposed by various researches for an automatic objective evaluation of stitched images is troublesome, or even impossible This situation is caused by significant differences between the most common types of distortions and those which may be found in stitched images. One of the methods recently proposed for quality assessment of stitched panoramic images, verified using the ISIQA database, is the Stitched Image Quality Evaluator (SIQE) proposed by the authors of this dataset [3] This method utilized a comparison of 36 features calculated for the constituent and stitched images, namely the eigenvalues determined for the covariance matrix where the covariances are computed for each patch for the pair of wavelet coefficients for a bivariate distribution determined from the Gaussian Mixture Model (GMM) and shape parameters of the Generalized Gaussian. The additional use of the variance of the local entropy and two additional features originating from the MIQM metric [9,10], leading to the extension of the idea initially verified in the paper [4], makes it possible to increase the correlations with subjective scores significantly, as presented in the further part of the paper
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
