Assessing Quality in Stereoscopic Videos

As the world becomes more crowded with cars and traffic, the number of accidents is on the rise. It’s a stark reality that around 1.3 million people lose their lives in traffic accidents every year due to various factors. Among those factors, one major cause is poor visibility due to weather conditions like fog, haze, rain, and snow. These are issues beyond our control.

To help tackle this problem, technology and systems known as Advanced Driver Assistance Systems (ADAS) are being developed. These systems aim to imitate how humans see and react to the world through our eyes. By merging the view from both the left and right, we can gain depth perception, which leads to a better viewing experience. This kind of technology can significantly improve safety for drivers.

However, creating high-quality 3D videos can be tricky. We require high-resolution cameras, ample storage space, and high-speed data transfer. Unfortunately, errors in capturing or displaying can reduce the overall viewing experience. That's why developing quality assessment models to judge the quality of these videos is essential.

#Types of Quality Assessments

Quality assessment methods can be classified into two categories: subjective and objective.

#Subjective Assessment

Here, real people watch and rate the quality of a video. Though accurate, this method is also quite time-consuming. It’s vital because, ultimately, we create videos for viewers, and their opinions are crucial benchmarks.

#Objective Assessment

This method provides automated predictions about a video's quality, usually based on algorithms that mimic human ratings. Objective assessments can be further classified into three types:

1. Full Reference (FR): Requires original video for comparison.
2. Reduced Reference (RR): Needs some info from the original video.
3. No Reference (NR): Doesn’t need any original reference to judge quality.

#The Challenge of Stereoscopic Videos

Stereoscopic videos (those that provide a 3D effect) present their own set of challenges. They blend depth information with standard images, leading to an enhanced quality of experience (QoE) for the viewer. However, producing these 3D videos presents vital requirements, such as high-quality equipment and careful viewing conditions.

Sometimes, the process of encoding or decoding may lead to errors, subsequently impacting the viewing experience. This highlights the necessity of having reliable quality assessment models for stereoscopic content.

#The Importance of Quality Assessment Models

While 2D quality assessment models are widely available, the field of 3D video quality assessment is still developing. Many researchers are working to improve how we measure video quality, but there’s still room for growth, especially in considering factors specific to stereoscopic content.

This article aims to examine both subjective and objective quality assessment methods for stereoscopic videos, focusing on how environmental factors like fog and haze affect viewer experience.

#Creating a Stereoscopic Video Dataset

A key part of this research was the development of a dataset containing various levels of fog and haze distortions. To achieve this, we produced 12 pristine stereoscopic videos and a whopping 360 distorted versions. These videos mimic real-world visibility issues to understand how they affect viewer perception.

To build our dataset, we selected some high-quality pristine videos and subjected them to different levels of fog and haze. This allowed us to analyze how these distortions impact the quality of the videos.

#Conducting a Subjective Study

Next, we needed to assess the quality of the videos we had created. We conducted a study where 24 participants watched our videos and rated them. They were asked to score the videos from 'Bad' to 'Excellent' based on their perceptions.

This subjective analysis is important because it gives us insights directly from viewers, helping us understand what makes a quality viewing experience.

#Quality Assessment Techniques

To analyze the quality more objectively, we proposed a model that doesn’t need any original video for comparison. This model processes the collected data using various methods to evaluate the quality of the stereoscopic content.

#Generating Cyclopean Frames

One clever technique involves creating what’s called cyclopean frames. These frames combine the left and right views into one image. By evaluating these combined images, we can gain insights into the quality of the 3D videos.

#Natural Scene Statistics Analysis

Next, we analyze the characteristics present in natural scenes in these videos. By examining various features of the video at multiple scales, we can better understand how distortions impact perceived quality.

#Using Statistical Modeling

We apply statistical models to evaluate the relationship between the pristine and distorted videos. This helps us distinguish the differences and determine how much the quality has changed.

#Results of the Study

After running our assessments, we found some interesting results. The proposed model performed consistently well across various Datasets, even against other established quality assessment methods. This suggests that our approach may provide a valuable tool for assessing the quality of stereoscopic videos.

#Conclusion

In conclusion, as video technology evolves, the need for effective quality assessment becomes ever more critical. The combination of subjective and objective methods enhances our understanding of video quality, particularly for stereoscopic content.

By creating a dataset that simulates visibility issues and developing assessment models, we aim to contribute to improving the quality of viewing experiences in the future.

Life is not just about seeing in two dimensions. Let’s make it 3D!