近期，美图公司与国科大联合在ECCV 2024上发布了一项创新性技术——BlazeBVD，该技术针对视频闪烁问题提供了通用的解决方案，旨在提升视频质量和观看体验。BlazeBVD的推出，标志着在视频处理领域，特别是针对视频闪烁的挑战，取得了一次重大的突破。
美图公司旗下的专业手机视频编辑工具Wink，凭借其独创的视频画质修复能力，已成为短视频生态中用户量持续攀升的佼佼者。Wink的画质修复功能之所以火爆，正是因为它精准地洞察了用户在视频创作中遇到的痛点，如画面模糊、噪点严重、画质低等问题。在美图影像研究院（MT Lab）的强力技术支持下，Wink已推出包括画质修复 – 高清、画质修复 – 超清、画质修复 – 人像增强、分辨率提升等功能。
此次，美图影像研究院（MT Lab）与国科大合作，提出了基于尺度时间均衡（STE）的盲视频去闪烁（BVD）新方法BlazeBVD。该方法针对视频闪烁场景，旨在处理光照闪烁退化未知的低质量视频，同时保持视频内容和色彩的完整性。BlazeBVD的核心在于引入了直方图辅助解决方案，利用图像直方图来描绘光亮和闪烁信息，通过平滑处理直方图，生成奇异帧集合、滤波光照图和曝光掩码图，实现快速、稳定的纹理恢复，尤其在光照波动、曝光过度或不足的情况下效果显著。
BlazeBVD的创新之处在于，它首次利用直方图降低BVD任务的学习复杂度，简化了处理视频数据的复杂性和资源消耗，同时结合了全局闪烁去除模块（GFRM）和局部闪烁去除模块，利用闪烁先验信息，不仅提高了视频的视觉稳定性，还提升了处理效率和质量。
BlazeBVD的发布，不仅为视频编辑和处理领域带来了新的可能，也体现了美图与国科大在人工智能和计算机视觉技术上的深度合作与创新。未来，随着技术的进一步发展和应用，BlazeBVD有望在视频修复、老电影修复、高速相机拍摄等领域发挥更大的作用，为用户提供更加优质和流畅的视频观看体验。
英语如下：
Headline: “美图 & 国科大联合发布创新视频去闪烁技术BlazeBVD”
Keywords: BlazeBVD, 美图, 国科大
News Content: Headline:美图与国科大联合在ECCV 2024上发布创新技术BlazeBVD，解决视频闪烁问题
In a recent development,美图 and the Chinese Academy of Sciences’ University of Science and Technology (国科大) jointly introduced BlazeBVD at ECCV 2024, a pioneering solution addressing the issue of video flickering. This innovation aims to enhance video quality and viewing experience. BlazeBVD’s launch signifies a significant breakthrough in the field of video processing, particularly in tackling the challenges posed by video flickering.
Wink, a professional mobile video editing tool under the美图 umbrella, has gained popularity due to its innovative video quality restoration capabilities, positioning itself as a standout in the short video ecosystem. The popularity of Wink’s quality restoration function is attributed to its precise understanding of users’ pain points in video creation, such as blurry images, severe noise, and low-quality videos. Supported by the powerful technical assistance of the美图 Image Research Institute (MT Lab), Wink has introduced features such as high-definition video quality restoration, ultra-high-definition video quality restoration, portrait enhancement, and resolution enhancement.
In collaboration with国科大, the美图 Image Research Institute (MT Lab) and the University of Science and Technology of China have proposed a new method, BlazeBVD, based on scale-time equilibrium (STE), for blind video flicker reduction (BVD). This method targets video flicker scenarios, dealing with low-quality videos where the source of flickering is unknown, while preserving the video’s content and color integrity. The core of BlazeBVD lies in the introduction of histogram-assisted solutions, using image histograms to depict luminance and flicker information. Through smoothing the histogram, it generates a collection of strange frames, a filtered light map, and an exposure mask, achieving rapid and stable texture recovery, especially effective in scenarios with fluctuating lighting, overexposure, or underexposure.
BlazeBVD’s innovation lies in its utilization of histograms to reduce the complexity of learning in the BVD task, simplifying the handling of video data and resource consumption. It combines a global flicker removal module (GFRM) and a local flicker removal module, leveraging flicker prior information to not only improve the visual stability of videos but also enhance processing efficiency and quality.
The release of BlazeBVD not only introduces new possibilities for video editing and processing but also highlights the depth of cooperation and innovation in artificial intelligence and computer vision technologies between美图 and 国科大. Looking ahead, as the technology continues to evolve and be applied, BlazeBVD is poised to play a significant role in video restoration, restoration of old films, and high-speed camera shooting, thereby providing users with a more premium and seamless video viewing experience.
【来源】https://www.jiqizhixin.com/articles/2024-07-23
Views: 2