Robust and Discriminative Feature Learning via Mutual Information Maximization for Object Detection in Aerial Images

Object detection in unmanned aerial vehicle(UAV)aerial images has become increasingly important in military and civil applications.General object detection models are not robust enough against interclass similarity and intraclass variability of small objects,and UAV-specific nuisances such as uncontrolledweather conditions.Unlike previous approaches focusing on high-level semantic information,we report the importance of underlying features to improve detection accuracy and robustness fromthe information-theoretic perspective.Specifically,we propose a robust and discriminative feature learning approach through mutual information maximization(RD-MIM),which can be integrated into numerous object detection methods for aerial images.Firstly,we present the rank sample mining method to reduce underlying feature differences between the natural image domain and the aerial image domain.Then,we design a momentum contrast learning strategy to make object features similar to the same category and dissimilar to different categories.Finally,we construct a transformer-based global attention mechanism to boost object location semantics by leveraging the high interrelation of different receptive fields.We conduct extensive experiments on the VisDrone and Unmanned Aerial Vehicle Benchmark Object Detection and Tracking(UAVDT)datasets to prove the effectiveness of the proposed method.The experimental results show that our approach brings considerable robustness gains to basic detectors and advanced detection methods,achieving relative growth rates of 51.0%and 39.4%in corruption robustness,respectively.Our code is available at https://github.com/cq100/RD-MIM(accessed on 2 August 2024).

Adaptive Boundary and Semantic Composite Segmentation Method for Individual Objects in Aerial Images

There are two types of methods for image segmentation.One is traditional image processing methods,which are sensitive to details and boundaries,yet fail to recognize semantic information.The other is deep learning methods,which can locate and identify different objects,but boundary identifications are not accurate enough.Both of them cannot generate entire segmentation information.In order to obtain accurate edge detection and semantic information,an Adaptive Boundary and Semantic Composite Segmentation method(ABSCS)is proposed.This method can precisely semantic segment individual objects in large-size aerial images with limited GPU performances.It includes adaptively dividing and modifying the aerial images with the proposed principles and methods,using the deep learning method to semantic segment and preprocess the small divided pieces,using three traditional methods to segment and preprocess original-size aerial images,adaptively selecting traditional results tomodify the boundaries of individual objects in deep learning results,and combining the results of different objects.Individual object semantic segmentation experiments are conducted by using the AeroScapes dataset,and their results are analyzed qualitatively and quantitatively.The experimental results demonstrate that the proposed method can achieve more promising object boundaries than the original deep learning method.This work also demonstrates the advantages of the proposed method in applications of point cloud semantic segmentation and image inpainting.

Two-stage image segmentation based on edge and region information

A two-stage method for image segmentation based on edge and region information is proposed. Different deformation schemes are used at two stages for segmenting the object correctly in image plane. At the first stage, the contour of the model is divided into several segments hierarchically that deform respectively using affine transformation. After the contour is deformed to the approximate boundary of object, a fine match mechanism using statistical information of local region to redefine the external energy of the model is used to make the contour fit the object's boundary exactly. The algorithm is effective, as the hierarchical segmental deformation makes use of the globe and local information of the image, the affine transformation keeps the consistency of the model, and the reformative approaches of computing the internal energy and external energy are proposed to reduce the algorithm complexity. The adaptive method of defining the search area at the second stage makes the model converge quickly. The experimental results indicate that the proposed model is effective and robust to local minima and able to search for concave objects.

LGNet:Local and global representation learning for fast biomedical image segmentation

Medical image segmentation plays a crucial role in clinical diagnosis and therapy systems,yet still faces many challenges.Building on convolutional neural networks(CNNs),medical image segmentation has achieved tremendous progress.However,owing to the locality of convolution operations,CNNs have the inherent limitation in learning global context.To address the limitation in building global context relationship from CNNs,we propose LGNet,a semantic segmentation network aiming to learn local and global features for fast and accurate medical image segmentation in this paper.Specifically,we employ a two-branch architecture consisting of convolution layers in one branch to learn local features and transformer layers in the other branch to learn global features.LGNet has two key insights:(1)We bridge two-branch to learn local and global features in an interactive way;(2)we present a novel multi-feature fusion model(MSFFM)to leverage the global contexture information from transformer and the local representational features from convolutions.Our method achieves state-of-the-art trade-off in terms of accuracy and efficiency on several medical image segmentation benchmarks including Synapse,ACDC and MOST.Specifically,LGNet achieves the state-of-the-art performance with Dice's indexes of 80.15%on Synapse,of 91.70%on ACDC,and of 95.56%on MOST.Meanwhile,the inference speed attains at 172 frames per second with 224-224 input resolution.The extensive experiments demonstrate the effectiveness of the proposed LGNet for fast and accurate for medical image segmentation.

Image segmentation by level set evolution with region consistency constraint

Image segmentation is a key and fundamental problem in image processing,computer graphics,and computer vision.Level set based method for image segmentation is used widely for its topology flexibility and proper mathematical formulation.However,poor performance of existing level set models on noisy images and weak boundary limit its application in image segmentation.In this paper,we present a region consistency constraint term to measure the regional consistency on both sides of the boundary,this term defines the boundary of the image within a range,and hence increases the stability of the level set model.The term can make existing level set models significantly improve the efficiency of the algorithms on segmenting images with noise and weak boundary.Furthermore,this constraint term can make edge-based level set model overcome the defect of sensitivity to the initial contour.The experimental results show that our algorithm is efficient for image segmentation and outperform the existing state-of-art methods regarding images with noise and weak boundary.

Flotation bubble image segmentation based on seed region boundary growing

Segmenting blurred and conglutinated bubbles in a flotation image is done using a new segmentation method based on Seed Region and Boundary Growing（SRBG）.Bright pixels located on bubble tops were extracted as the seed regions.Seed boundaries are divided into four curves：left-top,right-top,right-bottom, and left-bottom.Bubbles are segmented from the seed boundary by moving these curves to the bubble boundaries along the corresponding directions.The SRBG method can remove noisy areas and it avoids over- and under-segmentation problems.Each bubble is segmented separately rather than segmenting the entire flotation image.The segmentation results from the SRBG method are more accurate than those from the Watershed algorithm.

Watershed-based Image Segmentation with Region Merging and Edge Detection

The clustering technique is used to examine each pixel in the image which assigned to one of the clusters depending on the minimum distance to obtain primary classified image into different intensity regions. A watershed transformation technique is then employes. This includes: gradient of the classified image, dividing the image into markers, checking the Marker Image to see if it has zero points (watershed lines). The watershed lines are then deleted in the Marker Image created by watershed algorithm. A Region Adjacency Graph (RAG) and Region Adjacency Boundary (RAB) are created between two regions from Marker Image. Finally region merging is done according to region average intensity and two edge strengths (T1, T2). The approach of the authors is tested on remote sensing and brain MR medical images. The final segmentation result is one closed boundary per actual region in the image.

Medical Image Registration Based on Phase Congruency and Regional Mutual Information

In this paper, a new approach of muhi-modality image registration is represented with not only image intensity, but also features describing image structure. There are two novelties in the proposed method. Firstly, instead of standard mutual information （ MI ） based on joint intensity histogram, regional mutual information （ RMI ） is employed, which allows neighborhood information to be taken into account. Secondly, a new feature images obtained by means of phase congruency are invariants to brightness or contrast changes. By incorporating these features and intensity into RMI, we can combine the aspects of both structural and neighborhood information together, which offers a more robust and a high level of registration accuracy.

Desertification Detection in Makkah Region based on Aerial Images Classification

Desertification has become a global threat and caused a crisis,especially in Middle Eastern countries,such as Saudi Arabia.Makkah is one of the most important cities in Saudi Arabia that needs to be protected from desertification.The vegetation area in Makkah has been damaged because of desertification through wind,floods,overgrazing,and global climate change.The damage caused by desertification can be recovered provided urgent action is taken to prevent further degradation of the vegetation area.In this paper,we propose an automatic desertification detection system based on Deep Learning techniques.Aerial images are classified using Convolutional Neural Networks(CNN)to detect land state variation in real-time.CNNs have been widely used for computer vision applications,such as image classification,image segmentation,and quality enhancement.The proposed CNN model was trained and evaluated on the Arial Image Dataset(AID).Compared to state-of-the-art methods,the proposed model has better performance while being suitable for embedded implementation.It has achieved high efficiency with 96.47% accuracy.In light of the current research,we assert the appropriateness of the proposed CNN model in detecting desertification from aerial images.

SEED REGION SELECTION AND HOMOGENEITY CRITERION FOR DOORPLATE IMAGE SEGMENTATION IN MOBILE ROBOT NAVIGATION

Focused on the seed region selection and homogeneity criterion in Seeded Region Growing (SRG), an unsupervised seed region selection and a polynomial fitting homogeneity criterion for SRG are proposed in this paper. First of all, making use of Peer Group Filtering (PGF) techniques, an unsupervised seed region selection algorithm is presented to construct a seed region. Then based on the constructed seed region a polynomial fitting homogeneity criterion is applied to solve the concrete problem of doorplate segmentation appearing in the robot navigation along a corridor. At last, experiments are performed and the results demonstrate the effectiveness of the proposed algorithm.

An Image Segmentation Algorithm Based on a Local Region Conditional Random Field Model

To reduce the computation cost of a combined probabilistic graphical model and a deep neural network in semantic segmentation, the local region condition random field (LRCRF) model is investigated which selectively applies the condition random field (CRF) to the most active region in the image. The full convolutional network structure is optimized with the ResNet-18 structure and dilated convolution to expand the receptive field. The tracking networks are also improved based on SiameseFC by considering the frame relations in consecutive-frame traffic scene maps. Moreover, the segmentation results of the greyscale input data sets are more stable and effective than using the RGB images for deep neural network feature extraction. The experimental results show that the proposed method takes advantage of the image features directly and achieves good real-time performance and high segmentation accuracy.

Review of Theory and Methods of Image Segmentation

Image segmentation refers to the technique and process of partitioning a digital image into multiple segments based on image characteristics so as to extract the object of interest from it. It is a key step from image processing to image analysis. In the mid-1950s, people began to study image segmentation. For decades, various methods for image segmentation have been proposed. In this paper, traditional image segmentation methods and some new methods appearing in recent years were reviewed. Thresholding segmentation methods, region-based, edge detection-based and segmentation methods based on specific theoretical tools were introduced in detail.

Image Segmentation Using an Improved Watershed Algorithm

As watershed algorithm suffers from over-segmentation problem, this paper presented an efficient method to resolve this problem. First, pre-process of the image using median filter is made to reduce the effect of noise. Second, watershed algorithm is employed to provide initial regions. Third, regions are merged according to the information between the region and boundary. In the merger processing based on the region information, an adaptive threshold of the difference between the neighboring regions is used as the region merge criteria, which is based on the human visual character. In the merger processing on the boundary information, the gradient is used to judge the true boundary of the image to avoid merging the foreground with the background regions. Finally, post-process to the regions using mathematical morphology open and close filter is done to smooth object boundaries. The experimental results show that this method is very efficient.

New normalized nonlocal hybrid level set method for image segmentation

This article introduces a new normalized nonlocal hybrid level set method for image segmentation.Due to intensity overlapping,blurred edges with complex backgrounds,simple intensity and texture information,such kind of image segmentation is still a challenging task.The proposed method uses both the region and boundary information to achieve accurate segmentation results.The region information can help to identify rough region of interest and prevent the boundary leakage problem.It makes use of normalized nonlocal comparisons between pairs of patches in each region,and a heuristic intensity model is proposed to suppress irrelevant strong edges and constrain the segmentation.The boundary information can help to detect the precise location of the target object,it makes use of the geodesic active contour model to obtain the target boundary.The corresponding variational segmentation problem is implemented by a level set formulation.We use an internal energy term for geometric active contours to penalize the deviation of the level set function from a signed distance function.At last,experimental results on synthetic images and real images are shown in the paper with promising results.

A Novel Fast Algorithm of Mono Transition Region Determination with Gray Image

There is a transition region between objects and background in any gray image. Many valuable applications of image segmentation and edge detection based on transition region determination have been developed in recent years. But, the complexity of calculation for determining transition region is too high. It results in the very limitation of applications based on transition region. A new novel fast method for transition region determination is presented in this paper, which will reduce the complexity of calculation dramatically. Many experiments have showed that this algorithm is effective and correct and will lay a good foundation for applications based on transition region. Key words image segmentation - transition region - maximum point - efficient average gradient (EAG) CLC number TP 391.4 Biography: Zhang Ai-hua (1965-), male, Ph. D candidate, research direction: image processing.

AUTOMATIC SEGMENTATION OF HIPPOCAMPAL SUBFIELDS BASED ON MULTI-ATLAS IMAGE SEGMENTATION TECHNIQUES

The volume of hippocampal subfields is closely related with early diagnosis of Alzheimer's disease.Due to the anatomical complexity of hippocampal subfields,automatic segmentation merely on the content of MR images is extremely difficult.We presented a method which combines multi-atlas image segmentation with extreme learning machine based bias detection and correction technique to achieve a fully automatic segmentation of hippocampal subfields.Symmetric diffeomorphic registration driven by symmetric mutual information energy was implemented in atlas registration,which allows multi-modal image registration and accelerates execution time.An exponential function based label fusion strategy was proposed for the normalized similarity measure case in segmentation combination,which yields better combination accuracy.The test results show that this method is effective,especially for the larger subfields with an overlap of more than 80%,which is competitive with the current methods and is of potential clinical significance.

Improve the Nonparametric Image Segmentation with Narrowband Levelset and Fast Gauss Transformation

Nonparametric method based on the mutual information is an efficient technique for the image segmentation. In this method, the image is divided into the internal and external labeled regions, and the segmentation problem constrained by the total length of the region boundaries will be changed into the maximization of the mutual information between the region labels and the image pixel intensities. The maximization problem can be solved by deriving the associated gradient flows and the curve evolutions. One of the advantages for this method does not need to choose the segmentation parameter;another is not sensitive to the noise. By employing the narrowband levelset and Fast Gauss Transformation, the computation time is reduced clearly and the algorithm efficiency is greatly improved.

A framework of region-based dynamic image fusion

A new framework of region-based dynamic image fusion is proposed. First, the technique of target detection is applied to dynamic images (image sequences) to segment images into different targets and background regions. Then different fusion rules are employed in different regions so that the target information is preserved as much as possible. In addition, steerable non-separable wavelet frame transform is used in the process of multi-resolution analysis, so the system achieves favorable characters of orientation and invariant shift. Compared with other image fusion methods, experimental results showed that the proposed method has better capabilities of target recognition and preserves clear background information.

EFFICIENT IMAGE SEGMENTATION FOR SEMANTIC OBJECT GENERATION

This letter presents an efficient and simple image segmentation method for semantic object spatial segmentation. First, the image is filtered using contour-preserving filters. Then it is quasi-flat labeled. The small regions near the contour are classified as uncertain regions and are eliminated by region growing and merging. Further region merging is used to reduce the region number. The simulation results show its efficiency and simplicity. It can preserve the semantic object shape while emphasize on the perceptual complex part of the object. So it conforms to the human visual perception very well.

SOFT IMAGE SEGMENTATION BASED ON CENTER-FREE FUZZY CLUSTERING

Image segmentation remains one of the major challenges in image analysis.And soft image segmentation has been widely used due to its good effect.Fuzzy clustering algorithms are very popular in soft segmentation.A new soft image segmentation method based on center-free fuzzy clustering is proposed.The center-free fuzzy clustering is the modified version of the classical fuzzy C-means ( FCM ) clustering.Different from traditional fuzzy clustering , the center-free fuzzy clustering does not need to calculate the cluster center , so it can be applied to pairwise relational data.In the proposed method , the mean-shift method is chosen for initial segmentation firstly , then the center-free clustering is used to merge regions and the final segmented images are obtained at last.Experimental results show that the proposed method is better than other image segmentation methods based on traditional clustering.