Color mismatch in stereoscopic 3D (S3D) images can create visual discomfort and affect the performance of S3D image processing algorithms, e.g., for depth estimation. In this paper, we propose a new deep learning-based solution for the problem of color mismatch correction. The proposed solution consists of a multi-task convolutional neural network, where color correction is the primary task and correspondence estimation is the secondary task. For the training and evaluation of the proposed network, a new S3D image dataset with color mismatch was created. Based on this dataset, experiments were conducted showing the effectiveness of our solution.

Shooting a live-action immersive 360-degree experience, i.e. omnidirectional content (ODC) is a technological challenge as there are many technical limitations which need to be overcome, especially for capturing and post-processing in stereoscopic 3D (S3D). In this paper, we introduce a novel approach and entire system for stitching and color mismatch correction and detection in S3D omnidirectional content, which consists of three main modules: pre-processing, spherical color correction and color mismatch evaluation. The system and its individual modules are evaluated on two datasets, including a new dataset which will be publicly available with this paper. We show that our system outperforms the state of the art in color correction of S3D ODC and demonstrate that our spherical color correction module even further improves the results of the state of the art approaches.

This work proposes a method for geometric and color rectification of stereoscopic images. The method is based on the Optical Flow of every stereoscopic image pair and it follows global and local approximations for both rectifications. Although the method is automatic, it also allows the user to interact in order to obtain the best result. The method has been fully tested and it is suitable to be included in an industrial film postproduction environment.

In this paper we propose a method for noise reduction in image sequences, based on the optical flow, consisting in tracking each pixel’s position in the previous and following images. In some cases the optical flow imperfections can cause artifacts. To prevent them we also propose an improvement to the method based on the estimation of the imperfections of the optical flow. Using that estimation we adaptatively choose either a temporal or a spatial based noise reduction algorithm to be applied in different image zones. Our results show an important noise reduction, even with complex image sequences.

In this paper, we describe an automatic method for detecting and repairing non-repetitive damages in image sequences, caused by dust, fibers or local defects of the film emulsion. The method is a three frame window scheme based on the calculation of the optical flow (OF) relating adjacent frames and the first and the last frames of the sequence. The OF validity is checked in order to detect non-repetitive damage, and is later repaired using filtering and smooth blending of the damaged zones. The method works correctly for the set of tested image sequences providing perfect visual repairs of the damaged zones.

We present an optical flow based method for noise reduction in image sequences. To prevent artefacts caused by optical flow imperfections, we propose a method to estimate these imperfections. We use the estimation to adaptively choose either a temporal or a spatial based noise reduction algorithm to be applied in different image zones. Our results have shown that an important noise reduction can be achieved with the proposed method, without the drawbacks of the simpler methods. The method has provided important noise reductions even with complex image sequences.

In this paper we examine the performance advantages of using a GPU to execute the space variant Gaussian filtering. Our results show that the straightforward convolution GPU implementation obtains up to 8 times better performance than the best recursive algorithm (the Deriche’s filter) executed on a CPU, for useful maximum σ values. GPUs have turned out a useful option to obtain high execution performance, specially due to the emergence of high level languages for graphics hardware.

In this paper we describe an application of the optical flow in order to create a slow motion effect and frame rate conversion on streaming image sequences. Our results show that using the optical flow based method the image sequence shows much less artifacts than than in traditional interpolation or mixing methods.