Nearest neighbor search is a core operation in hyperspectral image processing tasks such as clustering and density-based analysis. In unsupervised settings, adaptive neighborhood definitions that do not rely on prior information are particularly valuable. Natural Neighbors (NaN) provide a parameter-free, data-driven neighborhood construction based on mutual nearest-neighbor relationships, but their practical use is hindered by the high computational and memory cost of repeated exact searches in large hyperspectral datasets. This paper investigates whether exact nearest neighbor search within the NaN framework can be replaced by approximate nearest neighbor (ANN) methods without significantly compromising neighborhood quality. Several ANN techniques, including HNSW, ANNOY, Product Quantization (PQ), PQ-IVF, and KD-tree, are integrated into the NaN algorithm. We propose an evaluation strategy that jointly assesses neighborhood quality and computational efficiency, combining recall, a modified silhouette index, runtime, and memory usage. Experiments on synthetic and real hyperspectral images show that approximate search can preserve the essential properties of Natural Neighbors with limited precision loss for the best-performing methods, while significantly improving scalability.

Modern imaging systems produce a great volume of image data. In many practical situations, it is necessary to compress them for faster transferring or more efficient storage. Then, a compression has to be applied. If images are noisy, lossless compression is almost useless, and lossy compression is characterized by a specific noise filtering effect that depends on the image, noise, and coder properties. Here, we considered a modern HEIF coder applied to grayscale (component) images of different complexity corrupted by additive white Gaussian noise. It has recently been shown that an optimal operation point (OOP) might exist in this case. Note that the OOP is a value of quality factor where the compressed image quality (according to a used quality metric) is the closest to the corresponding noise-free image. The lossy compression of noisy images leads to both noise reduction and distortions introduced into the information component, thus, a compromise should be found between the compressed image quality and compression ratio attained. The OOP is one possible compromise, if it exists, for a given noisy image. However, it has also recently been demonstrated that the compressed image quality can be significantly improved if post-filtering is applied under the condition that the quality factor is slightly larger than the one corresponding to the OOP. Therefore, we considered the efficiency of post-filtering where a block-matching 3-dimensional (BM3D) filter was applied. It was shown that the positive effect of such post-filtering could reach a few dB in terms of the PSNR and PSNR-HVS-M metrics. The largest benefits took place for simple structure images and a high intensity of noise. It was also demonstrated that the filter parameters have to be adapted to the properties of residual noise that become more non-Gaussian if the compression ratio increases. Practical recommendations on the use of compression parameters and post-filtering are given.

Visual information, such as images, is one of the most important parts of the digital realm. Its volume quickly increases due to better resolution of imagers and a large amount of images acquired each day. The acquired images should be stored on personal computers, servers, etc., and/or transferred via communication lines. A common way to reduce the data size is to apply compression techniques where lossy compression is often preferable because it is able to provide a significantly higher compression ratio compared to lossless compression. Meanwhile, it is necessary to control quality of lossy compressed images to avoid (minimize) negative consequences of introduced distortions. To solve this task, this paper proposes ways to predict mean squared error (MSE) of introduced distortions for better portable graphics (BPG) lossy compression. Such a prediction can be helpful in scenarios when an image has to be compressed in visually lossless manner or an appropriate compromise between the attained compression ratio (CR) and compressed image quality should be produced. It is shown that MSE is highly correlated not only with compression control parameter Q used in the BPG encoder, but with characteristics (complexity) of an image to be compressed. We present and compare several approaches to MSE prediction based on image fast preliminary analysis using local image activity and entropy. The practical recommendations for the MSE prediction are given.

Modern means of remote sensing (RS) in general and multispectral imaging in particular have experienced rapid development in recent years [...]

A lot of modern remote sensing images are multichannel and, due to this as well as to high resolution, they occupy quite a large space. This causes problems in their transfer and storage and leads to the necessity to apply compression where lossy compression is mainly used. The compressed images can be then processed in different ways where classification is a typical operation for which trained neural networks are widely used. Classifier performance depends on many factors including what are the images employed in training. We have earlier shown that if compressed images are planned to be classified, it is worth using just compressed images for training. However, images used for training and employed in classification can be obtained by different compression techniques. Hence, in this paper, we analyze and compare the results of using the same and different coders for training and classified images. It is demonstrated that the difference in classification accuracy is not large if one uses the same coder compressed data for training and classification or if the compression techniques are different. The largest difference has been observed for the situations when one coder is DCT-based and the other coder is wavelet-based.

The object of the study is the process of lossy compression of noisy images and their post-filtering. The subject of the study is the approach to efficient two-stage processing (compression and post-filtering) for better portable graphics (BPG) coder and prediction of its efficiency. The goal of the study is to analyze performance characteristics of the considered two-stage approach and to propose an approach to their prediction. Methods used: numerical simulation, regression, statistical analysis. Results obtained: 1) the considered approach advantage is that it is able to provide improvement of quality of compressed noisy image under condition that an image is compressed with compression ratio smaller than that one corresponding to optimal operation point; 2) the approach efficiency depends on several factors including noise intensity, image complexity, and filter type and parameters; 3) the main characteristics of the two-step procedure can be quite accurately predicted in advance and this allows offering useful information for decision undertaking on what value of the coder parameter to apply; 4) this leads to either improving the compressed and processed image quality compared to its original version or, at least, to avoiding quality degradation. Conclusions: based on the results of the study, it is worth 1) predicting performance characteristics for the two-stage processing; 2) adapting the processing to image complexity and noise intensity.

Lossy compression of remote-sensing images is a typical stage in their processing chain. In design or selection of methods for lossy compression, it is commonly assumed that images are noise-free. Meanwhile, there are many practical situations where an image or a set of its components are noisy. This fact needs to be taken into account since noise presence leads to specific effects in lossy compressed data. The main effect is the possible existence of the optimal operation point (OOP) shown for JPEG, JPEG2000, some coders based on the discrete cosine transform (DCT), and the better portable graphics (BPG) encoder. However, the performance of such modern coders as AVIF and HEIF with application to noisy images has not been studied yet. In this paper, analysis is carried out for the case of additive white Gaussian noise. We demonstrate that OOP can exist for AVIF and HEIF and the performance characteristics in it are quite similar to those for the BPG encoder. OOP exists with a higher probability for images of simpler structure and/or high-intensity noise, and this takes place according to different metrics including visual quality ones. The problems of providing lossy compression by AVIF or HEIF are shown and an initial solution is proposed. Examples for test and real-life remote-sensing images are presented.

Image registration aims at aligning two or more images in a common coordinate system. A common approach is to find a geometric transformation that explains the parameters of putative correspondences between images. Each correspondence measures the spatial translation between local patches of the registered images. The uncertainty of this translation is important for registration task allowing to weight each correspondence with respect to others. Measuring uncertainty is challenging especially for multimodal registration case and for anisotropic textures. In this paper, we develop further a learning-based patch matching approach that predicts the covariation matrix of the translation error for each correspondence. We identify the bias related to the orientation of anisotropic textures and propose additional data augmentation and a modified loss function to eliminate it. Using multimodal remote sensing images, we confirm that the proposed approach predicts the covariation matrix of translation errors for both isotropic and anisotropic textures with different orientations.

With a resolution improvement, the size of modern remote sensing images increases. This makes it desirable to compress them, mostly by using lossy compression techniques. Often the images to be compressed (or some component images of multichannel remote sensing data) are noisy. The lossy compression of such images has several peculiarities dealing with specific noise filtering effects and evaluation of the compression technique's performance. In particular, an optimal operation point (OOP) may exist where quality of a compressed image is closer to the corresponding noise-free (true) image than the uncompressed (original, noisy) image quality, according to certain criterion (metrics). In such a case, it is reasonable to automatically compress an image under interest in the OOP neighborhood, but without having the true image at disposal in practice, it is impossible to accurately determine if the OOP does exist. Here we show that, by a simple and fast preliminary analysis and pre-training, it is possible to predict the OOPs existence and the metric values in it with appropriate accuracy. The study is carried out for a better portable graphics (BPG) coder for additive white Gaussian noise, focusing mainly on one-component (grayscale) images. The results allow for concluding that prediction is possible for an improvement (reduction) in the quality metrics of PSNR and PSNR-HVS-M. In turn, this allows for decision-making about the existence or absence of an OOP. If an OOP is absent, a more "careful" compression is recommended. Having such rules, it then becomes possible to carry out the compression automatically. Additionally, possible modifications for the cases of signal-dependent noise and the joint compression of three-component images are considered and the possible existence of an OOP for these cases is demonstrated.

Characterization of remote sensing (RS) sensors noise is important for optimization and analysis of potential accuracy of subsequent image processing stages (such as filtering or compression, image registration accuracy or objects classification). Blind noise parameters estimation (BNPE) allows analyzing sensor noise directly from noisy images. BNPE is complementary to direct sensor calibration methods and is able to provide additional noise parameters measurements when calibration procedure is not accurate or cannot be performed. Recent BNPE methods show improved performance by utilizing learning-based approach. This paper aims at solving the problem of lack of training data with precise ground truth noise parameters needed for training BNPE models. We focus on learning-based approach using calibrated real data, mixture of synthetic and real data and novel semi-supervised training with uncalibrated real data. The results show that BNPE method training in semi-supervised manner performs better than model-based approach and comparable to supervised training. These results open up possibility of easy extension of training datasets for BNPE methods.

This study attempts to solve these issues associated with hyperspectral (HS) data, i.e., coarse spatial resolution and high volume, by understanding the effect of deep learning and traditional dimensionality reduction on super-resolved products generated from the recently launched PRecursore IperSpettrale della Missione Applicativa (PRISMA) HS mission. Four single-frame super-resolution (SR) algorithms have been used to super-resolve a 30 m PRISMA scene of Ahmedabad, India and generate 15 m spatial resolution images with both spatial and spectral fidelity. Iterative back projection (IBP) and sparse representation (SIS) are the best and worst-performing SR algorithms following a comparative assessment and validation protocol. Next, denoising autoencoders and PCT computed using singular and eigenvalue decompositions have been executed on the original PRISMA, IBP and SIS-based super-resolved datasets. The resulting low-dimensional representations have been assessed to preserve the original dataset's topology using label-independent Lee and Verleysen's co-ranking matrix and loss of quality measure. Findings suggest that autoencoders are computationally expensive and require a higher neighbourhood size than PCT and its variants to produce a high-quality encoding. These insights remain significant for urban information extraction as there are few direct comparative assessments between machine learning-based linear and non-linear data compression methods in earlier studies.

This chapter considers the lossy compression of three-channel images corrupted by additive white Gaussian noise. The better portable graphics (BPG) encoder that has three different operation modes is studied. As for other encoders, a noise filtering effect is observed and it influences the encoder performance. For certain conditions (image complexity and noise intensity), the so-called optimal operation point (OOP) may exist. The main property of OOP is that the quality of a compressed image (calculated with respect to the true image) is better than the quality of the corresponding noisy image according to a given metric (either conventional such as mean square error or some visual quality metric). Knowledge concerning the OOP existence and compression parameters in it for a given noisy image can be useful. In the case of OOP existence, it is reasonable to compress a given image in OOP or its neighborhood. If OOP does not exist, a more “careful” lossy compression is expedient. The problem is that in practice the true image is not available, and, thus, it is impossible to calculate full-reference metrics for compressed and noise-free images. Meanwhile, we show in this paper that it is possible to predict compression parameters for a given noisy image in advance. Moreover, this prediction is fast and accurate enough for a decision undertaking on what coder parameter to set to reach OOP. In addition, in this paper, we demonstrate that it is possible to predict the compression parameters not only in OOP but also in its neighborhood. This allows for undertaking more reliable decisions. Prediction opportunities are shown for two visual quality metrics and three typical modes of the BPG coder operation: 4:4:4, 4:2:2, and 4:2:0. One novel way of prediction accuracy improvement based on the joint use of two input parameters is shown.

Data stream partitioning is an important technique in data mining to analyze data streams in real-time. In this context, lots of data stream partitioning methods have been proposed. In the state of the art, most existing methods need to specify the number of classes before partitioning and/or introduce user-defined parameters for which the parameter values may differ for different data sets. In practice, it is difficult to determine the number of classes and the parameter values. Therefore, we propose in this paper an unsupervised and non-parametric method based on the Optimized Fuzzy C-Means algorithm. It has 2 steps. First is to partition a series of data chunks and then partition the intermediate classes formed before. The performance of the proposed algorithm is evaluated and compared with the recent state-of-the-art methods on hyperspectral image data sets.

Remote sensing (RS) digital images have a great variety of applications in solving real-world problems. Modern sensors provide this type of data of a very high resolution, which, in combination with a great number of acquired images, makes a problem of compressing RS images of particular importance. In this letter, discrete atomic compression (DAC) and a problem of its spatial complexity reduction are considered. This approach provides data compression and protection features in combination with such image representation that is ready for applying different artificial intelligence methods. For this reason, its application to image processing is relevant. Several modifications that provide reducing the spatial complexity of DAC are proposed, and their efficiency is analyzed. In particular, it is shown that using a block splitting procedure, it is possible to get a significant decrease in additional memory expenses without DAC's efficiency degradation in terms of lossy image compression.

Lossy compression of remote sensing data has found numerous applications. Several requirements are usually imposed on methods and algorithms to be used. A large compression ratio has to be provided, introduced distortions should not lead to sufficient reduction of classification accuracy, compression has to be realized quickly enough, etc. An additional requirement could be to provide privacy of compressed data. In this paper, we show that these requirements can be easily and effectively realized by compression based on discrete atomic transform (DAT). Three-channel remote sensing (RS) images that are part of multispectral data are used as examples. It is demonstrated that the quality of images compressed by DAT can be varied and controlled by setting maximal absolute deviation. This parameter also strictly relates to more traditional metrics as root mean square error (RMSE) and peak signal-to-noise ratio (PSNR) that can be controlled. It is also shown that there are several variants of DAT having different depths. Their performances are compared from different viewpoints, and the recommendations of transform depth are given. Effects of lossy compression on three-channel image classification using the maximum likelihood (ML) approach are studied. It is shown that the total probability of correct classification remains almost the same for a wide range of distortions introduced by lossy compression, although some variations of correct classification probabilities take place for particular classes depending on peculiarities of feature distributions. Experiments are carried out for multispectral Sentinel images of different complexities.

A method for predicting probability characteristics for algorithms of supervised classification of multichannel images is proposed. Implementation of quasi-Bayesian image recognition strategy in conditions of incomplete a priori information about a satellite image is considered. Total weighted probability of correct class recognition is taken as a criterion of effectiveness of decision rule. Weights are interpreted as estimates of a priori probability for occurrence of precedents of the corresponding classes. The procedure for estimating a priori probabilities of classes by Monte Carlo method based on the results of fuzzy classification of samples with fixed size is described; the empirical dependence of the relative error of estimates on the sample size is given.