Rectifying Irregular Boundaries Stereoscopic Images Using Deep Learning and Evolutionary Algorithms

Application of stereoscopic technology has increased significantly in several industries, including the media and entertainment sectors, medical imaging, and virtual reality. However, the overall viewing experience and the feeling of depth can be negatively impacted by uneven boundaries in stereoscopic images. To have a realistic and immersive stereoscopic experience, fixing these wavy edges is an essential process. This work present a distinctive approach for optimizing the rectangular bounds of stereoscopic images using a hybrid of deep learning and evolutionary algorithms. The suggested method uses a genetic algorithm to optimize the parameters of a convolutional neural network (CNN) to extract useful features for residual regression. Due to the lack of compatible datasets, a set of stereoscopic images was constructed from monocular images by calculating the depth of the objects using MiDaS pretrained depth estimation algorithm. Several metrics, including as mean squared error, peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and Fréchet inception distance (FID), were used to evaluate the performance of the suggested technique. The experimental findings show that the proposed method effectively optimizes rectangular borders with excellent accuracy, even for images with complicated shape and orientations. Various parameter settings were also tested, revealing that increasing the number of generations and population size contributes to improved performance. The combination of the deep learning algorithm, specifically the CNN, and the optimization algorithm, specifically the genetic algorithm (GA), leads to an increase in the accuracy of rectified images. Notably, the average PSNR reached 24.42, SSIM measured at 0.7793, and FID scored 18.26. The proposed method holds promise for a wide range of applications, including image cropping, object detection, and image segmentation, among others. By effectively optimizing rectangular boundaries in stereoscopic images, this approach enhances the quality and accuracy of various visual tasks, making it a valuable tool in multiple domains of study.