深度解析：图像裁剪算法——缝合裁剪（Seam Carving）原理与应用

需积分: 0 60 浏览量更新于2024-08-05 收藏 6.52MB PDF 举报

"谷正阳在中山大学的一次计算机视觉课程中深入探讨了'Seam Carving'这一主题。Seam Carving是一种内容感知的图像重缩放方法，它旨在解决网页图片在不同设备屏幕上适配时，如何保持关键内容完整的问题。原始的图片尺寸调整方式，如简单缩放或裁剪，可能会导致图像失真或重要部分丢失。在讲座中，谷正阳首先介绍了Seam Carving的基本概念，它通过逐行（或列）删除对图像内容影响最小的像素，形成所谓的"缝"（seam），以此来实现图片的精确重缩放。这种方法避免了直接缩放可能带来的视觉扭曲，同时尽可能保留画面的视觉连贯性和完整性。为了深入了解这个算法，谷正阳参考了论文[2]，这是一篇首次提出Seam Carving技术的文章，以及其他相关的研究。这些论文详细阐述了如何选择最不重要的像素进行删除，以及如何利用图像特征（如纹理、颜色一致性等）来确定哪些像素属于缝，从而确保重缩放后的图片既能减小尺寸，又能保持关键信息。讲座进一步探讨了Seam Carving的优势，例如： 1. 适应性：对于需要在各种屏幕尺寸上显示的图片，Seam Carving能提供更自然的适应效果，避免了传统方法可能导致的图像变形。 2. 内容保留：相比于简单的裁剪，Seam Carving能保留边缘附近的细节，使得图像在缩放后仍保持原有的视觉吸引力。 3. 计算效率：虽然理论上可能需要遍历整个图像，但通过有效的算法优化，Seam Carving可以在可接受的时间内完成图像重缩放。 4. 应用场景：除了网页图片适配，Seam Carving还被应用在视频编辑、图像压缩等领域，用于保持视频帧之间的连续性和画面质量。总结来说，谷正阳在这次讲座中不仅分享了Seam Carving的基本原理，还展示了其在实际图像处理中的实用性和优越性，这对于理解和掌握内容感知的图像处理技术具有重要意义。"

Lecture Node for Seam Carving

Zhengyang Gu Sun Yat-sen University guzhy3@mail2.sysu.edu.cn

Abstract

The lecture gave us some basic concepts about seam

carving, which are very intriguing. Therefore, I read the

papper[2] which rst introduce seam carving and some

other related pappers, in order to dig deeper into this al-

gorithm.

1. Introduction

Image retargeting is to change the size of pictures on a

website when the website is accessed by dierent divices

with various screen sizes. The simplest way is scaling and

cropping1. However, scaling may distort important contents

in pictures. For example, the cat1 becomes obviously dis-

torted after scaling. And cropping may remove important

contents at the periphery of pictures. For instance, some

of the tentacles near the edges of the picture1 are removed.

Therefore, content-aware image resizing approaches that

can maintain important contents when resizing pictures is

in great demand.

Seam carving is a content-aware image resizing appo-

rach. The basic idea of it is to remove the most unimpor-

tant pixels in a seam which is intuitively a line. Therefore, it

can eliminate redundance and keep consistency of pictures

at the same time. For example, the picture2 is resized by

seam carving, while the cat is not distorted and the tentacles

are not removed.

2. Background

There are some other content-aware image resizing meth-

ods mentioned by the paper[2]. For example, use top down

methods like face detectors[10] or bottom down methods

like visual saliency methods[7] to build saliency map. And

then based on the saliency map, use intelligent cropping

methods to resize the image[9]. Another paper[4] pays at-

tention to users’ perception on the adapted images. Their

method automatically detect and transmit the most impor-

tant part of the image to devices.

And the paper[2] also mentioned other areas where seams

are used. For instance, the paper[1] nds the best seams

to combine parts of a set of pictures into a single compos-

ite picture. And the computation of seams can be done by

(a) Original

(b) Scaling (c) Cropping

Figure 1: The original picture shows a cat with some tenta-

cles. The picture is scaled and cropped.

Figure 2: Seam carving

other ways, including Dijkstra’s shortest path algorithm[6]

and graph cuts[3].

下载后可阅读完整内容，剩余8页未读，立即下载

kdbshi

粉丝: 739
资源: 298

深度解析：图像裁剪算法——缝合裁剪（Seam Carving）原理与应用

"计算机组成与结构：lecture 3-计算机功能和互联视角

计算机视觉：立体视觉算法详解与深度计算

"2020春-lecture81: 计算机体系结构第八讲的内容总结

ecs-summer-school-vision-lecture:ECS暑期学校计算机视觉讲座的交互式幻灯片

Data Compression Techniques - Lecture 4 - Integer Codes 2 - University of Helsinky - Slides (DCT2015-Lecture4)-计算机科学

Curves and Surfaces - Lecture Notes for Geometry 1 - Henrik Schlichtkrull - University of Copenhagen (2011)-计算机科学

【计算机视觉】Lecture9：立体视觉算法 计算机视觉.pdf

Lecture-notes-for-Machine-Learning:Lecture notes for Machine Learning (机器学习讲义)

Jeff-Dean-s-Lecture-for-YC-AI

283-lecture0-ACG计算机图形学-19秋review190909B.pdf

最新资源

【计算机视觉】Lecture9：立体视觉算法计算机视觉.pdf