Significant disparities between the features of natural images and those inherent to histopathological images make it challenging to directly apply and transfer pre-trained models from natural images to histopathology tasks. Moreover, the frequent lack of annotations in histopathology patch images has driven researchers to explore self-supervised learning methods like mask reconstruction for learning representations from large amounts of unlabeled data. Crucially, previous mask-based efforts in self-supervised learning have often overlooked the spatial interactions among entities, which are essential for constructing accurate representations of pathological entities. To address these challenges, constructing graphs of entities is a promising approach. In addition, the diffusion reconstruction strategy has recently shown superior performance through its random intensity noise addition technique to enhance the robust learned representation. Therefore, we introduce H-MGDM, a novel self-supervised Histopathology image representation learning method through the Dynamic Entity-Masked Graph Diffusion Model. Specifically, we propose to use complementary subgraphs as latent diffusion conditions and self-supervised targets respectively during pre-training. We note that the graph can embed entities' topological relationships and enhance representation. Dynamic conditions and targets can improve pathological fine reconstruction. Our model has conducted pretraining experiments on three large histopathological datasets. The advanced predictive performance and interpretability of H-MGDM are clearly evaluated on comprehensive downstream tasks such as classification and survival analysis on six datasets.

International audience

Fisheye lenses are renowned for their capacity to capture incredibly broad perspectives, often reaching up to 180 degrees. Their versatility extends beyond computer vision tasks and encompasses various fields. While applying computer vision techniques directly to fisheye images can yield suboptimal results, this article aims to introduce an uncomplicated and straightforward adaptation of the well-known Bayesian based particle filter. Through a few minor adjustments, we will demonstrate the potential to enhance the particle filter’s performance when dealing with such images, particularly for tracking purposes. In this study, we investigate the applicability of particle filters based on features such as: color space, Local Binary Pattern, Histogram of Oriented Gradients, and their combinations. Eventually, experiments and evaluations were carried out on hand annotated real videos with a top view fisheye camera in motion.

Existing filtering techniques fail to precisely detect adjacent line features in multi-scale applications. In this paper, a new filter composed of a bi-Gaussian and a semi-Gaussian kernel is proposed, capable of highlighting complex linear structures such as ridges and valleys of different widths, with noise robustness. Experiments have been performed on a set of both synthetic and real images containing adjacent line features. The obtained results show the performance of the new technique in comparison to the main existing filtering methods.

For the last twenty years, the mortality of honeybee colonies has been increasing and is now a major concern for biologists. There have been a number of hypotheses put forth in the literature, including hives' infestation with parasites like Varroa destructor, the appearance of new predators like Asian hornets (Vespa Velutina) or small hive beetles and widespread use of pesticides. In addition, climate and environmental changes are modifying the honeybee's interaction with its ecosystem. Last works suggest that this mortality could be due not to a single reason, but to a convergence of stress factors. This article presents the development of an experimental platform dedicated to track the honeybee colony (Apis mellifera) health in parallel with the quantification of the main environmental stressors. Our main objective is to provide biologists with a paraphernalia that is primarily based on motion analyses with different sensors inside and outside the hive for the study of bees' behavior at both individual (the bee) as well as a colony (the superorganism) scales. The aim is to create a reference dataset of behaviors associated with environmental factors in both normal and stressful environments.

The goal of human tracking is to detect people in a scene and assign them a unique identifier that the tracker will follow across multiple frames. Our tracker, FORT, implements deep learning solutions such as, YOLOv7 for detection, ResNeXt-50 for feature extraction and re-identification and an adapted Kalman filter for tracking. The goal is to present a real time tracking solution for the complex environment of top-view, fisheye images. The proposed solution is then compared with the BoT-SORT and StrongSORT trackers on a custom fisheye Multiple Object Tracking (MOT) Challenge dataset.

Pedestrian detection has large relevance to the understanding of static and moving scenes of video sequences. The increasing demand for safety and security of people has resulted in more research on intelligent visual surveillance in a wide range of applications, such as moving human detection. With the great success of deep learning methods, researchers decided to switch from traditional methods based hand-crafted feature extractors to recent deep learning-based techniques in order to detect and track people. In this work, the topic of person detection with a Top-view moving fisheye camera is addressed. Although the fisheye camera is a useful tool for video monitoring, most of object detection techniques, with (or without) deep learning, concern classical perspective cameras. However, due to the distortions of fisheye images, we are expected to have higher requirements and challenges on the pedestrian detection using this device. In this paper, we propose an end-to-end learning people detection method based on YOLOv3 detector that detects people using oriented bounding boxes. The proposed model customizes the traditional YOLOv3 for the detection of oriented bounding boxes, by regressing the angle of each bounding box using a periodic loss function. With rotation bounding box prediction, our approach is efficient, reaching 98,1% of true detection. The proposed method is evaluated on a new available dataset where rotated bounding boxes represent annotations from several fisheye videos: https://partage.imt.fr/index.php/s/nytmFqiq8jaztkX

Most of embedded cameras use a single sensor to capture images through a color filter. They produce special images with only one color component per pixel. Missing data are usually estimated through a demosaicking process, but this takes undesirable computation time and may generate undesirable color artifacts. Many embedded systems under real-time constraints could use this type of camera for purposes like edge detection. In this paper, a new edge detection method is proposed for the computation of partial derivative images. This algorithm is tested on a large set of synthetic images where edge ground truth is unquestionable. The exploitation of the raw data of images allows not only to drastically reduce computational time, but also to get edge detection results even more precise than using certain demosaicking-based methods.

Figure of merit represents an expression characterizing the performance of an algorithm. In edge detection assessment, it corresponds to a supervised evaluation by quantifying differences between a reference edge map and a candidate, computed by a performance measure/criterion. This paper introduces a new normalized supervised edge detection evaluation measure which provides an overall evaluation of the quality of a contour map, by taking into account the amount of false positives, false negatives and degrees of shifting. Finally, an objective assessment performed by varying the hysteresis thresholds on contours of real images shows that the new measure outperforms six other compared normalized methods, in term of evaluation and visualization of the detected contours.

Vehicle classification is an important topic which is still under research consideration because of its role in road surveillance, security system, traffic monitoring, and accident prevention. In this paper, we propose a deep learning model for vehicles classification using the Convolutional Neural Networks (CNN) integrated with a statistical moments layer. We referred to the model as ICNN. As an additional layer, the moments layer extracts statistical moments features from the feature maps obtained from convolutions layers. The moments layer is fed the fully-connected classifier of the network which is fine-tuned to get better results. Our Integrated CNN model (ICNN) achieves 97.1% accuracy compared to the most popular algorithms used in this field such as K Nearest Neighbour (KNN), and Support Vector Machine (SVM), which known as good tools for object classification.

This article deals with color images steganalysis based on machine learning. The proposed approach enriches the features from the Color Rich Model by adding new features obtained by applying steerable Gaussian filters and then computing the co-occurrence of pixel pairs. Adding these new features to those obtained from Color-Rich Models allows us to increase the detectability of hidden messages in color images. The Gaussian filters are angled in different directions to precisely compute the tangent of the gradient vector. Then, the gradient magnitude and the derivative of this tangent direction are estimated. This refined method of estimation enables us to unearth the minor changes that have occurred in the image when a message is embedded. The efficiency of the proposed framework is demonstrated on three stenographic algorithms designed to hide messages in images: S-UNIWARD, WOW, and Synch-HILL. Each algorithm is tested using different payload sizes. The proposed approach is compared to three color image steganalysis methods based on computation features and Ensemble Classifier classification: the Spatial Color Rich Model, the CFA-aware Rich Model and the RGB Geometric Color Rich Model.

Abstract In recent years, information security has received a great deal of attention. To give an example, steganography techniques are used to communicate in a secret and invisible way. Digital color images have become a good medium for digital steganography because of their easy manipulation as carriers via Internet, e‐mails, or used on websites. The main goal of steganalysis is to detect the presence of hidden messages in a digital media. The proposed method is a further extension of the authors' previous work: steganalysis based on color feature correlation and machine learning classification. Fusing features with those obtained from color‐rich models allows increasing the detectability of hidden messages in the color images. Our new proposition uses two types of features, computed between color image channels. The first type of feature reflects local Euclidean transformations, and the second one reflects mirror transformations. These geometric measures are obtained by the sine and cosine of gradient angles between all the color channels. Features are extracted from co‐occurrence correlation matrices of measures. We demonstrate the efficiency of the proposed framework on three steganography algorithms designed to hide messages in images represented in the spatial domain: S‐UNIWARD, WOW, and Synch‐HILL. For each algorithm, we applied a range of different payload sizes. The efficiency of the proposed method is demonstrated by the comparison with the previous authors work and the spatial color‐rich model and color filter array‐aware features for steganalysis. Copyright © 2016 John Wiley & Sons, Ltd.

Digital images, especially color images, are very widely used, as well as traded via Internet, e-mail and posting on websites. Images have a large size which allows embedding secret messages of large size, so they are a good medium for digital steganography. The main goal of steganalysis is to detect the presence of hidden messages in digital media. In this paper, we propose a steganalysis method based on color feature correlation and machine learning classification. Fusing features with features obtained from color-rich models allows to increase the detect ability of hidden messages in the color images. Our novel proposition uses the correlation between different channels of color images. Features are extracted from the channel correlation and co-occurrence correlation. In this work, all stego images are created with a range of different payload sizes using two steganography S-UNIWARD and WOW algorithms. To validate the proposed method, his efficiency is demonstrated by comparison with color rich model steg analysis.

Crest line extraction has always been a challenging task in image processing and its applications. It is possible to detect ridges and valleys in images using second order filters. In order to estimate crest lines of variable widths, a multi-scale analysis of the image is required. In this paper we propose a new ridge/valley detection method in images based on the difference of rotating Gaussian semi filters adapted in a multi-scale process. Due to the directional filters, we obtain a new ridge/valley anisotropic detector enabling very precise detection of ridge/valley of varied widths. Moreover, as the detector filters compute the two directions of crest lines, even highly bended crest lines are correctly extracted. Numerical comparisons with other oriented Gaussian filters and results on real images validate the interest of this method.

In this paper, we present a new method for removing texture in images using a smoothing rotating filter. From this filter, a bank of smoothed images provides pixel signals able to classify a pixel as a texture pixel, a homogenous region pixel or an edge pixel. Then, we introduce a new method for anisotropic diffusion which controls accurately the diffusion near edge and corner points and diffuses isotropically inside textured regions. Several results applied on real images and a comparison with anisotropic diffusion methods show that our model is able to remove the texture and control the diffusion.