Reflection symmetry aware image retargeting
Introduction
Reflection symmetry is the most frequently occurring type of symmetry in natural objects and man-made objects. While retargeting an image, there have been attempts to preserve the translation symmetry while summarizing an image [1]. However, retargeting of images containing reflective symmetric objects is still an unsolved problem. The standard seam carving method and its variations, in order to preserve the geometry of an object, first generate an energy map. This energy map has high values at the pixels corresponding to the object to be preserved. However, the energy map itself is not symmetric, i.e., it may not have the same value at the reflective symmetric pixels. Therefore, if a seam passes through one side of a reflective symmetric object, then it is not necessary that another seam will pass through the pixels which are mirror reflections of the pixels of the seam . Furthermore, finding a reflective symmetric energy map can be a computationally prohibitive task [2]. Therefore, instead of finding a reflective symmetric energy map, we only find the reflection symmetry axis of a symmetric object which is a computationally tractable problem.
In this work, we propose a novel and efficient image retargeting approach which preserves the objects exhibiting reflection symmetry. We find the reflection symmetry axis from the image. However, just the symmetry axis is not sufficient to give complete information about the location and the shape of the object in the image. We use object proposal which is better in describing a symmetry region given the symmetry axis and the symmetric point pairs. In order to preserve reflection symmetry, we propose a symmetry aware seam carving approach. We first find a seam using the traditional seam carving approach. Now, in order to preserve the reflection symmetry, we find the reflective seam corresponding to this seam using the symmetry axis and the object proposal we have estimated. Then, we remove or add (for shrinking or expanding the input image) both the seams. However, we observe that a seam after reflection about the symmetry axis may not follow the criteria to be a seam. The reason is that the symmetry axis may not always be a vertical line. Therefore, integer pixel locations may not be integers after reflection. Also, a reflective vertical (horizontal) seam does not follow an 8-connected path of pixels in the image from top to bottom (left to right), containing only one pixel in each row (column) of the image. We propose an optimization based approach in order to make the reflection of a seam also to be a seam. After reflection and optimization of a seam, we propagate an optimized reflected seam outside the symmetry region. The primary contributions of the proposed work are as follows.
- 1.
We introduce a symmetry aware image retargeting operator that preserves the reflection symmetry present in an image. We find the reflection symmetry region using the symmetry axis, the symmetric point pairs and object proposals avoiding the need for object segmentation or symmetry map.
- 2.
We reflect the seam in the symmetry region about the symmetry axis. For preserving the symmetry region, we propose an optimization framework to make the reflected seam satisfy the seam constraint. We further use propagation to complete the optimized seam in the non-symmetry region.
- 3.
We propose a novel evaluation framework to validate the performance of the reflection symmetry aware image retargeting.
The rest of the paper is structured as follows. In Section 2, we present a brief overview of the symmetry detection and the image retargeting algorithms. Section 3 explains the proposed approach of the reflection symmetry aware image retargeting. The evaluation framework and the results are discussed in Section 4. We conclude the paper in Section 5 with the future scope.
Section snippets
Related work
The problems of symmetry detection and image retargeting have been extensively studied in the field of computer vision and computer graphics.
Symmetry detection. The problem of detecting reflection symmetry axis of a reflective symmetric object present in a given image has been an active problem [3], [4], [5], [6], [7], [8]. Loy and Eklundh in [9] used SIFT keypoints [10] of the symmetric object to find the symmetry axis. They proposed a Hough transform based voting method where each pair of
Proposed approach
A natural or a synthetic image contains both symmetry and non-symmetry regions. We first estimate the energy map, the symmetry axis, and the multiple object proposals of an image using the algorithm proposed in [40], [12] and [41], respectively. The object proposal generation algorithm finds a set of rectangular proposals indicating the location of objects with some probability. However, the object proposal algorithm is not symmetry aware. It does not provide the best object proposal which
Results and evaluation
To validate the performance of the proposed approach, we conduct experiments on two datasets: Detecting Symmetry in the Wild (DSW) reflection symmetry dataset [3] and RetargetMe dataset [42]. The DSW reflection symmetry dataset consists of 194 images (100 images from training dataset and 94 images from testing dataset) each containing a single symmetric object. The images consist of natural as well as synthetic scenes. The images for which symmetry axis is not detected using the symmetry axis
Conclusion
In this work, we have addressed the problem of retargeting for images exhibiting reflection symmetry. With the aid of object proposals, we have detected the symmetry region using the global symmetry axis and the symmetric point pairs. After reflecting a seam passing through the symmetry region, we have proposed an optimization framework for the reflected seam to satisfy the seam constraint. We have propagated the optimized reflected seam for it to be extended over the image size. Also, we have
References (48)
- et al.
Multiple instance subspace learning via partial random projection tree for local reflection symmetry in natural images
Pattern Recognit.
(2016) - et al.
Seam searching-based pixel fusion for image retargeting.
IEEE Trans. Circuits Syst. Video Tech.
(2015) - et al.
Resizing by symmetry-summarization
ACM Trans. Graphics (TOG)
(2010) - et al.
Symmmap: estimation of 2-d reflection symmetry map and its applications
Proceedings, ICCV Workshop on Detecting Symmetry in the Wild
(2017) - et al.
2017 ICCV challenge: detecting symmetry in the wild
The IEEE International Conference on Computer Vision (ICCV) Workshops
(2017) - et al.
Computational Symmetry in Computer Vision and Computer Graphics
(2010) - et al.
Symmetry in 3d geometry: extraction and applications
Computer Graphics Forum
(2013) - et al.
Symmetry detection from realworld images competition 2013: summary and results
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops
(2013) - et al.
Symmetry detection competition: a summary of how the competition is carried out
Proc. IEEE CVPR Workshop Symmetry Detection Real World Images
(2011) - et al.
Detecting approximate reflection symmetry in a point set using optimization on manifold
IEEE Trans. Signal Process.
(2019)
Detecting symmetry and symmetric constellations of features
European Conference on Computer Vision
Distinctive image features from scale-invariant keypoints
Int. J. Comput. Vis.
Reflection symmetry detection via appearance of structure descriptor
European Conference on Computer Vision
Wavelet-based reflection symmetry detection via textural and color histograms
The IEEE International Conference on Computer Vision (ICCV) Workshops
Finding mirror symmetry via registration and optimal symmetric pairwise assignment of curves
The IEEE International Conference on Computer Vision (ICCV) Workshops
Learning-based symmetry detection in natural images
European Conference on Computer Vision
Deepskeleton: learning multi-task scale-associated deep side outputs for object skeleton extraction in natural images
IEEE Trans. Image Process.
Image retargeting using mesh parametrization
IEEE Trans. Multimed.
A shape-preserving approach to image resizing
Computer Graphics Forum
Scale and object aware image retargeting for thumbnail browsing
Computer Vision (ICCV), 2011 IEEE International Conference on
A line-structure-preserving approach to image resizing
2012 IEEE Conference on Computer Vision and Pattern Recognition
Patch-based image warping for content-aware retargeting
IEEE Trans. Multimed.
A structure-aware axis-aligned grid deformation approach for robust image retargeting
Multimed. Tools Appl.
Image retargeting via Beltrami representation
IEEE Trans. Image Process.
Cited by (13)
A Comprehensive Review on Content-Aware Image Retargeting: From Classical to State-of-the-art Methods
2022, Signal ProcessingCitation Excerpt :Therefore, it is possible to remain the content inside the cropping window intact and discard the rest of the image. Recently, CAIR methods [1–94] have been proposed to solve these problems, which protect the important content of the image during the image retargeting process and minimize image distortion as much as possible. Each CAIR method usually consists of two main steps: 1) determining the importance of image pixels by extracting an importance map, and 2) performing an image resizing method to obtain the retargeted image.
Deep learning-based importance map for content-aware media retargeting
2024, Multimedia Tools and ApplicationsSeamless Image Editing for Perceptual Size Restoration Based on Seam Carving
2024, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Content-aware image retargeting technique and iterated function system: frameworks, applications, and possible future advancements
2023, Multimedia Tools and ApplicationsAnalysis of seam carving technique: limitations, improvements and possible solutions
2023, Visual ComputerPerformance analysis of seam diversion based image retargeting technique based on edge detection operators
2023, Multimedia Tools and Applications