谁能解释一下hq2x算法的原理？
我不知道hq2x算法的详细步骤。虽然我已经从网站下载了源代码，但我仍然无法理解源代码中的步骤。谁能帮我解释一下如何逐步执行该算法？谢谢你！
I do not know the detail steps of hq2x algorithm. Although I have downloaded the source code from website, I still cannot understand the steps from the source code. Can anyone help me to explain how to do the algorithm step by step? Thank you!
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
它同时查看多个像素（中心像素及其邻居）。使用对比度将组划分为亮像素和暗像素（将其视为 1 位位图）。然后它使用它来识别这些像素形成的图案（例如对角线）并输出该图案的预定义放大版本，并使用源像素进行着色。
例如，如果您注意到相邻像素形成该形状：
那么您可以输出它的高分辨率版本：
用于识别图案的区域是重叠的以保证连续性（或者以另一种方式思考它，它会查看更多像素以获取上下文，例如区分对角线和圆角）。
由于具有模式识别功能，hq2x 仅适用于高对比度卡通图像。对于具有大量模糊边缘和抗锯齿线的自然图像，算法无法注意到模式，最终只是对图像进行插值。
It looks at several pixels at once (a central pixel and its neighbors). Uses contrast to divide the group into light and dark pixels (think of it as 1-bit bitmap). Then it uses this to recognize a pattern that these pixels form (e.g. a diagonal line) and outputs a predefined, enlarged version of that pattern, colorized with source pixels.
e.g. if you notice that neighboring pixels form that shape:
then you can output higher-res version of it:
The areas for recognizing patterns are overlapping for continuity (or another way to think about it that it looks at more pixels for context, e.g. to distinguish diagonal line from rounded corner).
Because of pattern recognition, hq2x works well only on high-contrast cartoony images. With natural images, with lots of blurry edges and anti-aliased lines the algorithm fails to notice patterns and ends up simply interpolating the image.