Improving the quality of images synthesized by discrete cosines transform – regression based method using principle component analysis

  Purpose: Different views of an individuals’ image may be required for proper face recognition.   Recently, discrete cosines transform (DCT) based method has been used to synthesize virtual   views of an image using only one frontal image. In this work the performance of two different   algorithms was examined to produce virtual views of one frontal image.   Materials and Methods: Two new methods, based on neural networks (NN) and principle   component analysis (PCA) were used to make virtual views of an image. The results were   compared with those of the DCT-based method. Two distance metrics, i.e. mean square error   (MSE) and structural similarity index measure (SSIM), were used to measure and compare image   qualities. About 400 data were used to evaluate the performance of the new proposed methods.   Results: The neural networks fail to improve the quality of virtually produced images. However,   principle component analysis improved the quality of the synthesized images about 3%.   Conclusion: Principle component analysis is better than both DCT-based and neural network   methods for synthesizing virtual views of an image.