blocks|key|155292|text|你能解释一下你是如何计算渐变方向的吗？我相信您已经将每个4x4窗口分组，并计算了每个这样的窗口内渐变的方向。但所使用的sobel操作符的大小为5x5。这显然会导致一些重叠。你能解释一下吗？|type|unstyled|depth|inlineStyleRanges|entityRanges|data|155293|entityMap^0|0^^$0|@$1|2|3|4|5|6|7|D|8|@]|9|@]|A|$]]|$1|B|3|-4|5|6|7|E|8|@]|9|@]|A|$]]]|C|$]]

Could you please explain how you've computed orientation of gradients? I believe you've grouped each 4x4 window and computed the orientation of gradients inside each such window. But the sobel operator used has size 5x5. This obviously result in some overlap. Could you explain on this?

I am trying to implement a image matching algorithm based on gradient orientation matching. The main algo contains from the following steps:

<a href="https://i.stack.imgur.com/fvZ53.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/fvZ53.png" alt="enter image description here"></a>

<ol>
<li>convert the image to polar coordinates:</li>
</ol>

<a href="https://i.stack.imgur.com/m7VBl.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/m7VBl.png" alt="enter image description here"></a>

<ol start="2">
<li>calculate gradients using sobel operator:

Xgrad = cv2.Sobel(gr,cv2.CV_64F,1,0,ksize=5)</li>
</ol>

<a href="https://i.stack.imgur.com/nEm1X.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/nEm1X.png" alt="enter image description here"></a>

<pre><code>Ygrad = cv2.Sobel(gr,cv2.CV_64F,0,1,ksize=3)
</code></pre>

<a href="https://i.stack.imgur.com/dndHN.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/dndHN.png" alt="enter image description here"></a>

3) calculate orientation of gradient and binarize it. :

<a href="https://i.stack.imgur.com/3SKZL.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/3SKZL.png" alt="enter image description here"></a>

Now I can compare images using last features map with ignorance to rotation ans smal changes. 

But what I have found that this algo detects rotations of the same image with rotation rather purely. I have build a test image with circles to test this algo:

<a href="https://i.stack.imgur.com/dRBM1.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/dRBM1.png" alt="enter image description here"></a>

and rotated it to 10 grads

<a href="https://i.stack.imgur.com/YJamQ.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/YJamQ.png" alt="enter image description here"></a>

here are polar conversions:

<a href="https://i.stack.imgur.com/VMjvM.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/VMjvM.png" alt="enter image description here"></a>

<a href="https://i.stack.imgur.com/7yoVi.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/7yoVi.png" alt="enter image description here"></a>

and gradient orientation masks: as You se here is already a lot of noise on gradient matrix. and it brokes matching algo. 

<a href="https://i.stack.imgur.com/1D7xg.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/1D7xg.png" alt="enter image description here"></a>
<a href="https://i.stack.imgur.com/mWXss.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/mWXss.png" alt="enter image description here"></a>

and it's best passing difference mask: the whole line areas are marked as not matched. Small gaussian blurring on different steps are not helping at all. I dont know why. 

<a href="https://i.stack.imgur.com/p7QcC.png" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/p7QcC.png" alt="enter image description here"></a>

Update:

Gradient calculation:

<pre><code>gx = cv2.Sobel(gr,cv2.CV_64F,1,0,ksize=1)
gy = cv2.Sobel(gr,cv2.CV_64F,0,1,ksize=1)
blurredgx = cv2.GaussianBlur(gx,(11,3),1)
blurredgy = cv2.GaussianBlur(gy,(11,3),1)
magnitude, angle = cv2.cartToPolar(blurredgx, blurredgy)
</code></pre>

Calculating gradient orientation (HOG like) poor accuracy

翻译质量差，导致语言生硬或混乱。

没有提供实际的解决方法或示例。

解答不清晰，无法理解或解决问题。

页面排版不美观，阅读体验差。

文章

问答

视频

教程

学习中心

腾讯云实验室

直播

竞赛

腾讯云代码分析专区

腾讯iOA零信任安全管理系统专区

腾讯云架构师技术同盟交流圈

腾讯云数据库专区

腾讯云智能顾问专区

腾讯云原生专区

腾讯混元专区

腾讯云TCE专区

腾讯云Lighthouse专区

腾讯云HAI专区

腾讯云Edgeone专区

腾讯云存储专区

腾讯云智能专区

腾讯轻联专区 

腾讯云开发专区

TAPD专区

腾讯轻量云游戏服专区

EdgeOne AI 安全实战专区

腾讯云最具价值专家

腾讯云架构师技术同盟

腾讯云创作之星

腾讯云开发者先锋

腾讯云AI代码助手

云原生构建

TAPD 敏捷项目管理

Cloud Studio

SDK中心

API中心

命令行工具

功能1上新10个字符

功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符。

功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符。

功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符

功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符

功能4上新

文章&问答评论现已支持表情

全新交互，全新视觉，新增快捷键、悬浮工具栏、高亮块等功能并同时优化现有功能，全面提升创作效率和体验

社区富文本编辑器全新改版！诚邀体验～ 

精选全网热门MCP server，让你的AI更好用 🚀

💥开发者 MCP广场重磅上线！

涵盖代码开发、场景应用、自动测试全流程，助你从零构建专属AI助手

一站式MCP教程库，解锁AI应用新玩法

聚焦“写作效率、视觉美观与运行性能”三方面进行全面升级，为您提供更高效、稳定的创作环境

社区富文本&Markdown编辑器全新改版上线，欢迎大家体验!

诚挚邀请您参与本次调研，分享您的真实使用感受与建议。您的反馈至关重要，感谢您的支持与参与！

社区新版编辑器体验调研

本文尝试实现一种基于梯度方向匹配的图像匹配算法。主要的algo包括以下步骤：​​将图像转换为极坐标：​​使用sobel算子计算梯度：Xgrad = cv2.Sobel(gr，cv2.CV_64F,1,0，ksize=5)​​Ygrad = cv2.Sobel(gr,cv2.CV_64F,0,1,ksize=3)​​3)计算梯度的方向，并进行二值化。：​​现在，我可以比较图像使用最后的功能，地图与无

问计算梯度方向(HOG样)精度不高
EN

回答 1

Stack Overflow用户

社区

活动

圈层

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问计算梯度方向(HOG样)精度不高EN