Fig 6 - uploaded by Walid Barhoumi
Content may be subject to copyright.
Source publication
Mosaicing is connecting two or more images and making a new wide area image with no visible seam-lines. Several algorithms
have been proposed to construct mosaics from image sequence where the camera motion is more or less complex. Most of these
methods are based either on the interest points matching or on theoretical corner models. This paper des...
Contexts in source publication
Context 1
... invariant with respect to translation (Fig. 7), rotation ( Fig. 8) and scale factor (Fig. 9). In order to be validated, these couples must thereafter verify the constraint of the relative position [1]. We start by focusing on four couple of regions (⊂Λ) verifying a spatial likeness between quadrilaterals formed by their cen- ters of gravity (Fig. 6). This returns to the checking of the equations system (3) and of the orientation similarity bet- ween the two ...
Context 2
... the most widely adopted method for the treatment of external sources of error. RANSAC is a convenient tool to aim refinement since it considers outliers detection prior to the homogra- phy estimation. The application of RANSAC ameliorates the homography estimate robustness, by keeping only the set of feature matches which are loosely consistent (Fig. 16). Since RANSAC failed when the fraction of outliers is too great [4], we opted for its use as the last step of the interest points matching procedure ...
Context 3
... Fig. 26, we recorded the same number N in the case of a rotation around the optical axis. This was done by applying a gradually increasing rotation angle to an image of the input pair. We deduced that below a rotation angle value equals to π/8, the use of the zncc score (without regions matching) does not permit to match any point of interest. ...
Similar publications
Our purpose was to check the potential ability of oversampling as a method for computed tomography axial resolution improvement. The method of achieving isotropic and fine resolution, when the scanning system is characterized by anisotropic resolutions is proposed. In case of typical clinical system the axial resolution is much lower than the plana...
The paper presents a method of noise reduction in CT images by the curvelet transform. Noise affects the ability to visualize pathologic qualities and the living tissues structure in CT. Noise in CT images depends on the amount of discrete x-ray photons reaching the detector. In the CT images, noise is responsible for visibility reduction the low c...
It is well known that, in CT reconstruction, Maximum A Posteriori (MAP) reconstruction based on a Poisson noise model can be well approximated by Penalized Weighted Least Square (PWLS) minimization based on a data dependent Gaussian noise model. We study minimization of the PWLS objective function using the Gradient Descent (GD) method, and show th...
Citations
... Extracting the optimal seamline is necessary for achieving a seamless mosaic in high-resolution satellite imagery, where numerous objects are identifiable in detail (Li et al., 2019;Zagrouba et al., 2009). To efficiently extract seamlines in highresolution satellite images, we apply the Dijkstra's shortest path algorithm. ...
High-resolution satellite imagery has a limitation in terms of coverage area. This limitation presents challenges for extensive-scale analysis at regional or national levels. To maximize the utility of high-resolution satellite imagery, the implementation of image mosaicking techniques is essential. In this paper, we have developed seamline extraction techniques and relative geometric correction optimized for high-resolution satellite imagery. Ultimately, we proposed a multi-strip image mosaicking method for KOMPSAT-3A (Korea Multi-Purpose Satellite-3A) images. We applied the Dijkstra's shortest path algorithm to efficiently extract seamlines. we also performed image registration based on feature matching and homography transformation to correct the relative geometric errors between input images. We conducted experiments with our methods using 29 scenes from KOMPSAT-3A L1G data. The results indicated high relative geometric accuracy, with an average error of 1.63 pixels. Furthermore, we were able to obtain high-quality seamless mosaic images. Our proposed method is expected to enhance the utility of KOMPSAT-3A imagery for large-scale environmental and urban analysis and to provide more accurate and comprehensive data.
... Traditional mosaicing methods commonly use low-level features, i.e. pointbased mosaicing and HoG based mosaicing [70,138,74]. However, traditional mosaicing methods do not perform well due to poor medical image quality, e.g. ...
... Kanazawa et al. [70] extract 100 feature points generated from the Harris operator to find the correlation between two images. Zagrouba et al. [138] extract both Harris points and the semantic segmentation as multi-level features for correlation learning. Kourogi et al. [74] first generated the pseudo-motion vector on each pixel, and then estimate the affine motion parameters by using pixel-wise motion vector matching. ...
In the past ten years, with the help of deep learning, especially the rapid development of deep neural networks, medical image analysis has made remarkable progress. However, how to effectively use the relational information between various tissues or organs in medical images is still a very challenging problem, and it has not been fully studied. In this thesis, we propose two novel solutions to this problem based on deep relational learning. First, we propose a context-aware fully convolutional network that effectively models implicit relation information between features to perform medical image segmentation. The network achieves the state-of-the-art segmentation results on the Multi Modal Brain Tumor Segmentation 2017 (BraTS2017) and Multi Modal Brain Tumor Segmentation 2018 (BraTS2018) data sets. Subsequently, we propose a new hierarchical homography estimation network to achieve accurate medical image mosaicing by learning the explicit spatial relationship between adjacent frames. We use the UCL Fetoscopy Placenta dataset to conduct experiments and our hierarchical homography estimation network outperforms the other state-of-the-art mosaicing methods while generating robust and meaningful mosaicing result on unseen frames.
... Most image stitching methods use the information of the overlapping area of images to estimate the two-dimensional transformation between two images, such as homography and affine. The estimated two-dimensional transform is used to align the image [1,15,20,22]. However, the single homography cannot fit the complex scene well, and it can only get better stitching result under ideal conditions (the scene is a single plane or the camera optical center is fixed, that is, there is no parallax). ...
Image stitching aims to stitch multiple images with overlapping areas into a high-resolution image with natural appearance, no ghosting and no seam, so as to obtain a wide field of vision. And the stitching is required to be as fast as possible. In this article, we discuss the loss of foreground object caused by the seam-cut method for image stitching. Moreover, we propose an improved seam-cut method based on superpixel to solve this problem. The proposed method uses the matching information of feature points and superpixel segmentation to divide the scene into foreground and background. By adjusting the penalty value of foreground superpixels in the superpixel-based seam-cut method, the foreground object can avoid the cutting of optimal seam. The experimental results demonstrate that our method will not cause the ghosting of foreground object or the loss of foreground object in the stitching result in the scene with large parallax. In this method, the foreground object will be preserved completely.
... Image registration uses the extraction of characteristic features [20], such as lines, segment points, curves, corners, etc., used for image matching. They are based on twodimensional (2D) primitives [21] extraction and matching on overlapped images. The bottleneck of these methods is that they are computationally intensive [20]. ...
... This method emphasizes information theoretic concepts and can also be viewed as a statistical dependency measure between compared images using the histogram of the images [65]. ...
Intensive research has been done on image mosaic algorithms to improve the field of view through generated image mosaics. However, their usage varies from one field to another due to the challenges faced by image acquisition platforms. Moreover, the current imagery software packages used are computationally intensive to be used in real-time applications and are not economically affordable. Those that are open-source are limited due to the less amount of data used to test their mosaicing algorithms’ performance. Therefore, detailed knowledge of appropriate mosaic algorithms suitable for real-time applications is needed to produce mosaics with less computational time and efficient feature point detection. A comprehensive survey that categorizes existing mosaic algorithms’ adoption in various fields has not been done to the best of our knowledge. Firstly, we provide a comparison of the strengths, weaknesses, and uniqueness of the image mosaic algorithms across different fields, with emphasis on challenging issues, limitations, performance criteria, and mechanisms. Furthermore, this paper provides an up-to-date review of image mosaic algorithms in various domains as used in different fields. We further classify the usage of image mosaic algorithms based on the following domains: spatial, frequency, and combined spatial and frequency as used in agriculture, environmental monitoring, and medical imaging. In addition, an analysis was carried out on one of the promising algorithms based on improved SIFT with the aim of improvement. We then proposed an improved SIFT algorithm, which was then evaluated with an open-source algorithm and commercial software using structural similarity index measure (SSIM) and mosaicing computational times for mosaic accuracy and processing efficiency, respectively. Our approach demonstrated a significant improvement of more than 10% average on the mosaicing computational times for the five datasets used. Its mosaicing accuracy was found to be relatively within an acceptable range of above 90% averagely.
... For homography transformations, a significant step forward was using robust matching of invariant features (e.g., SIFT [11]) under a geometry model using RANSAC [12], as proposed by Brown and Lowe [21], [22], and followed on by many others. Multiple researchers (e.g., [23], [24], [25]) opted for using the Harris-Laplacian detector [26] to detect feature points, and some (e.g., [27], [28]) found advantages to using speeded up robust features (SURF) [29]. To evaluate matches within the RANSAC framework, in addition to the nearestneighbour based methods proposed originally, Zhao et al. [23] considered normalized correlation, and De Cesare et al. [24] proposed entropy and mutual information based measures. ...
We address fast image stitching for large image collections while being robust to drift due to chaining transformations and minimal overlap between images. We focus on scientific applications where ground truth accuracy is far more important than visual appearance or projection error, which can be misleading. For common large-scale image stitching use cases, transformations between images are often restricted to similarity or translation. When homography is used in these cases, the odds of being trapped in a poor local minimum and producing unnatural results increases. Thus, for transformations up to affine, we cast stitching as minimizing reprojection error globally using linear least squares with a few, simple constraints. For homography, we observe that the global affine solution provides better initialization for bundle adjustment compared to an alternative that initializes with a homography-based scaffolding, and at lower computational cost. We evaluate our methods on a very large translation dataset with limited overlap, as well as four drone datasets. We show that our approach is better compared to alternative methods such as MGRAPH in terms of computational cost, scaling to large numbers of images, and robustness to drift. We also contribute ground truth datasets for this endeavor.
... For homography transformations, a significant step forward was using robust matching of invariant features (e.g., SIFT [5]) under a geometry model using RANSAC [6], as proposed by Brown and Lowe [12], [13], and followed on by many others. Multiple researchers (e.g., [14], [15], [16]) opted for using the Harris-Laplacian detector [17] to detect feature points, and some (e.g., [18], [19]) found advantages to using speeded up robust features (SURF) [20]. To evaluate matches within the RANSAC framework, in addition to the nearest-neighbour based methods proposed originally, Zhao et al. [14] considered normalized correlation, and De Cesare et al. [15] proposed entropy and mutual information based measures. ...
We address fast image stitching for large image collections while being robust to drift due to chaining transformations and minimal overlap between images. We focus on scientific applications where ground truth accuracy is far more important than visual appearance or projection error, which can be misleading. For common large-scale image stitching use cases, transformations between images are often restricted to similarity or translation. When homography is used in these cases, the odds of being trapped in a poor local minimum and producing unnatural results increases. Thus, for transformations up to affine, we cast stitching as minimizing reprojection error globally using linear least squares with a few, simple constraints. For homography, we observe that the global affine solution provides better ini-tialization for bundle adjustment compared to an alternative that initializes with a homography-based scaffolding, and at lower computational cost. We evaluate our methods on a very large translation dataset with limited overlap, as well as four drone datasets. We show that our approach is better compared to alternative methods such as MGRAPH in terms of computational cost, scaling to large numbers of images, and robustness to drift. We also contribute ground truth datasets for this endeavor.
... For homography transformations, a significant step forward was using robust matching of invariant features (e.g., SIFT [5]) under a geometry model using RANSAC [6], as proposed by Brown and Lowe [12], [13], and followed on by many others. Multiple researchers (e.g., [14], [15], [16]) opted for using the Harris-Laplacian detector [17] to detect feature points, and some (e.g., [18], [19]) found advantages to using speeded up robust features (SURF) [20]. To evaluate matches within the RANSAC framework, in addition to the nearest-neighbour based methods proposed originally, Zhao et al. [14] considered normalized correlation, and De Cesare et al. [15] proposed entropy and mutual information based measures. ...
Method details are available in the manuscript.<br
... For homography transformations, a significant step forward was using robust matching of invariant features (e.g., SIFT [5]) under a geometry model using RANSAC [6], as proposed by Brown and Lowe [12], [13], and followed on by many others. Multiple researchers (e.g., [14], [15], [16]) opted for using the Harris-Laplacian detector [17] to detect feature points, and some (e.g., [18], [19]) found advantages to using speeded up robust features (SURF) [20]. To evaluate matches within the RANSAC framework, in addition to the nearest-neighbour based methods proposed originally, Zhao et al. [14] considered normalized correlation, and De Cesare et al. [15] proposed entropy and mutual information based measures. ...
Method details are available in the manuscript.<br
... In the high-level feature algorithms, one part of the image (contours) is identified as feature. In the low-level feature algorithms, keypoints or corners are identified as features [44,45]. Typically, low-level feature algorithms are used more often for the image mosaicing process than high-level feature algorithms [2,23,29]. ...
In this paper, a new method for image mosaicing (image stitching) is introduced based on Scale Invariant Feature transform (SIFT). One of the main drawbacks of SIFT is the redundancy of the extracted keypoints, which leads to lower image mosaicing quality. Recently, a new method called Redundant Keypoint Elimination (RKEM) was presented to remove these redundant features, and enhance image registration performance. Despite the applicability of RKEM, its threshold value is considered the same in all parts of the image. This characteristic leads to inappropriate removal of keypoints due to the fact that distribution of keypoints in the high-detailed region is denser than the low-detailed ones. This paper proposes a new method to improve RKEM called Clustered RKEM (CRKEM) which is based on keypoints distribution. Moreover, in this paper a new blending algorithm is proposed based on a Gaussian-weighted function. In the proposed blending method, the Gaussian function is proposed based on the mean and variance of the pixels in the overlapped region of images to be mosiaced. In comparison with the classical methods, the experimental results confirm the superiority of the proposed method in image mosaicing as well as to image registration and matching.
