A Registration Method for Historical Maps Based on Self-Supervised Feature Matching

The scope of this study is related to the classification of covid-19 pneumonia which uses X-ray images of the lungs as input and is then classified into 3 categories, namely normal (healthy), pneumonia, and covid-19 pneumonia. The classification process uses a deep learning approach, which is becoming a trend nowadays because its performance is better than conventional machine learning approaches. The image dataset in this study is a collection of X-ray images of the lungs consisting of 1,525 normal images, 1,525 pneumonia images, and 1,525 Covid-19 pneumonia images. These images vary in size, but all images are larger than 1000x1000 pixels. The large size of the image makes the training time long and the computational resources required are large. Deep learning machines need to reduce the size of the image to make the process more effective. However, changes in image size may affect classification performance. Therefore, we conducted experiments in this study to determine the effect of image size changes on the classification performance of COVID-19 pneumonia using a deep learning approach. We use the Convolutional Neural Network method, which uses 22 layers with 5 convolution layers, for classification. The experimental results show that the size of the input image affects the classification performance of covid-19 pneumonia. This study provides important information that larger image sizes do not always result in better classification performance. In addition, the image size that is too small causes very low classification performance. The performance of the classification of covid-19 pneumonia using the CNN method shows the most optimal results at an input image size of 500x500 pixels.

An iterative and subsequent proximation method to map historical crop information with satellite images

To solve the low accuracy of image feature matching in horticultural robot visual navigation, an innovative and effective image feature matching algorithm was proposed combining the improved Oriented FAST and Rotated BRIEF (ORB) and Lucas–Kanade (LK) optical flow algorithm. First, image feature points were extracted according to the adaptive threshold calculated using the Michelson contrast. Then, the extracted feature points were uniformed by the quadtree structure, which can reduce the calculated amount of feature matching, and the uniform ORB feature points were roughly matched to estimate the position of the feature points in the matched image using the improved LK optical flow. Finally, the Hamming distance between rough matching points was calculated for precise matching. Feature extraction and matching experiments were performed in four typical scenes: normal light, low light, high texture, and low texture. Compared with the traditional algorithm, the uniformity and accuracy of the feature points extracted by the proposed algorithm were enhanced by 0.22 and 50.47%, respectively. Meanwhile, the results revealed that the matching accuracy of the proposed algorithm increased by 14.59%, whereas the matching time and total time decreased by 39.18% and 44.79%, respectively. The proposed algorithm shows great potential for application in the visual simultaneous localization and mapping (V-SLAM) of horticultural robots to achieve higher accuracy of real-time positioning and map construction.

We developed a computer-aided diagnosis (CADx) method for classification between benign nodule, primary lung cancer, and metastatic lung cancer and evaluated the following: (i) the usefulness of the deep convolutional neural network (DCNN) for CADx of the ternary classification, compared with a conventional method (hand-crafted imaging feature plus machine learning), (ii) the effectiveness of transfer learning, and (iii) the effect of image size as the DCNN input. Among 1240 patients of previously-built database, computed tomography images and clinical information of 1236 patients were included. For the conventional method, CADx was performed by using rotation-invariant uniform-pattern local binary pattern on three orthogonal planes with a support vector machine. For the DCNN method, CADx was evaluated using the VGG-16 convolutional neural network with and without transfer learning, and hyperparameter optimization of the DCNN method was performed by random search. The best averaged validation accuracies of CADx were 55.9%, 68.0%, and 62.4% for the conventional method, the DCNN method with transfer learning, and the DCNN method without transfer learning, respectively. For image size of 56, 112, and 224, the best averaged validation accuracy for the DCNN with transfer learning were 60.7%, 64.7%, and 68.0%, respectively. DCNN was better than the conventional method for CADx, and the accuracy of DCNN improved when using transfer learning. Also, we found that larger image sizes as inputs to DCNN improved the accuracy of lung nodule classification.

Recent advances have enabled the extraction of vectorized features from digital historical maps. To fully leverage this information, however, the extracted features must be organized in a structured and meaningful way that supports efficient access and use. One promising approach is question answering (QA), which allows users—especially those unfamiliar with database query languages—to retrieve knowledge in a natural and intuitive manner. In this project, we developed a GeoQA system by integrating a spatio-temporal knowledge graph (KG) constructed from historical map data with large language models (LLMs). Specifically, we have defined the ontology to guide the construction of the spatio-temporal KG and investigated workflows of two different types of GeoQA—factual and descriptive. Additional data sources, such as historical map images and internet search results, are incorporated into our framework to provide extra context for descriptive GeoQA. Evaluation results demonstrate that the system can generate answers with a high delivery rate and a high semantic accuracy. To make the framework accessible, we further developed a web application that supports interactive querying and visualization.

Current work in combining multirate filtering techniques with a three-source (low pass, edge, and texture) image model designed to increase control over perceptual edge quality is described. The model employs multirate subband filtering to code the low-pass source efficiently and to reconstruct the image from the sources. Spline modeling can be used to exploit the two-dimensional connectivity of the edge source. Parameters of the edge profile (edge-source values along the perpendicular direction) are chosen so that coding errors are not too perceptually objectionable. Using a two-level octave band frequency configuration and without coding the sources, the SNR is 57.0 dB, perceptually indistinguishable from the original. >

VSN is different from traditional WSN that collect and transmit images/video instead of scalar data to the sink. Images provide richer information than scalar data that enable VSN to be a promising technology on security monitoring and surveillance. Image transmission in VSN is expansive due to its large image size and significant energy consumption. One possible way to reduce the image size is to aggregate the received images on the sensor nodes before transmission. By eliminating the redundant image, image aggregation in VSN could transmit the overlapped region of the images only once to reduce the total energy consumption. However, image aggregation requires image processing at the aggregator node that incurs additional node processing energy. In this paper, we study the energy consumption tradeoff between node image processing and image aggregation. We derive the image aggregation energy consumption model that considers the image processing and image transmission. From the simulation results, image aggregation scheme does help lowering the energy consumption. As compared to the non-aggregation scheme, image aggregation scheme consumes only 46% of the energy consumption. In addition, we also find that the threshold for the number of aggregated nodes. Above this threshold, increasing the number aggregated nodes does not help on reducing the energy consumption. This could help determining the number of aggregated nodes under different degree of aggregation to achieve minimum energy consumption.

Embedding the owner image is the fundamental concept in identifying the ownership of the QR code. However, the image larger than the percentage of error correction, embedded in the QR code, could not be read. This research try to increase the size of the embedding image by investigating the change of encoding the pad characters and the adjustment of intensity of light. The largest image size that can be embedded and read through general QR code reader, with different errors correction levels, is found. The results are compared with other research on embedding an image in except the finder patternof QR code and our embedding images size are larger in all the correction levels.

Medical images are often used for educational, analytical, and medical diagnostic purposes. Medical image data requires large amounts of storage on computers. Three types of codecs, namely Haar, Daubechies, and Biorthogonal, were used in this study. This study aims to find the best wavelet method of the three tested wavelet methods (Haar, Daubechies, and Biorthogonal). This study uses medical images representing USG and CT-scan images as testing data. The first test is carried out by comparing the threshold ratio. Three threshold values are used, namely 30, 40, and 50. The second test looks for PSNR values with different thresholds. The third test looks for a comparison of the rate (image size) to the PSSR value. The final test is to find each medical image's compression and decompression times. The first compression ratio test results on both medical images showed that CT scan images on Haar and Biorthogonal wavelets were the best, with an average compression ratio of 40.76% and a PSNR of 33.77. The PSNR obtained is also getting more significant for testing with a larger image size. The average compression time is 0.52 seconds, and the decompression time is 2.27 seconds. Based on the test results, this study recommends that the Daubechies wavelet method is very good for compression, which is 0.51 seconds, and the Biorthogonal wavelet method is very good for medical image decompression, which is 1.69 seconds.

An efficient MPEG-2 to MPEG-4 video transcoder is presented in this paper. We consider the transcoding from high quality and bit rate MPEG-2 video with larger image size (e.g. 4CIF/4SIF, CIF) to lower quality and bit rate MPEG-4 video with smaller image size (e.g. CIF, QCIF). First, the transcoder needs to down-sample the input MPEG-2 video. Since the motion vectors carried by the MPEG-2 stream will be reused in the transcoding process, they are sub-sampled besides the frame pixels, and the coding mode for each down-sampled macroblock is examined. A new rate control method is proposed to convert the high bit rate MPEG-2 video to the low bit rate MPEG-4 counterpart. The proposed rate control scheme adjusts the frame rate and the frame quantization step size simultaneously according to the channel bandwidth to achieve a good temporal-spatial quality tradeoff. Due to the reuse of motion vectors, key frames (i.e. I and P frames) cannot be skipped to maintain the prediction sequential order, while some B frames containing less temporal information may be skipped in transcoding to save the bit rate. Skipped B frames can be reconstructed at the decoder to ensure the full frame rate playback. The TM7 quadratic Rate-Qtz model is adopted in the proposed rate control scheme to calculate the re-quantization step size from the given target bit. Simulations show that the proposed MPEG-2 to MPEG-4 video transcoder with rate control can out-perform the basic MPEG transcoder that adjusts the re-quantization step size at a constant frame rate. The complexity of the proposed transcoder is low so that it can be used in real-time applications.

Mass loss of glaciers in the Chinese Altai was detected using geodetic methods based on topographical maps (1959), the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) (2000), and the Advanced Space-borne Thermal Emission and Reflection Radiometer (ASTER) stereo images (2008). The results indicate that a continued and accelerating shrinkage has occurred in the Chinese Altai Mountains during the last 50 years, with mass deficits of 0.43 ± 0.02 and 0.54 ± 0.13 m a−1 water equivalent (w.e.) during the periods 1959–1999 and 1999–2008, respectively. Overall, the Chinese Altai Mountains have lost 7.06 ± 0.44 km3 in ice volume (equivalent to −0.43 ± 0.03 m a−1 w.e.) from 1959–2008. The spatial heterogeneity in mass loss was potentially affected by comprehensive changes in temperature and precipitation, and had a substantial correlation with glacier size and topographic settings. Comparison shows that in the Chinese Altai Mountains glaciers have experienced a more rapid mass loss than those in the Tianshan and northwestern Tibetan Plateau (TP), and the mass balance of glaciers was slightly less negative relative to those in the Russian Altai, Himalaya, and southern TP.

From “red” to green? A look into the evolution of green spaces in a post-socialist city

Objects without prominent textures pose challenges for an automatic 3D model reconstruction and feature point matching. Such objects are common in many industrial applications such as metal defect detection, archeological applications of photogrammetry, and 3D object reconstruction from infrared imagery. Most of the common feature point descriptors fail to match local patches in featureless regions of an object. Various kind of textures requires different feature descriptors for high-quality image matching. Hence, automatic low-textured 3D object reconstruction using Structure from Motion (SfM) methods is challenging. Nevertheless, such reconstruction is possible with the aid of a human operator. Deep learning-based descriptors have outperformed most of common feature point descriptors recently. This paper is focused on the development of a new conditional generative adversarial auto-encoder (GANcoder) based on the deep learning. We use a coder-decoder architecture with four convolutional and four deconvolutional layers as a staring point for our research. Our main contribution is a generative adversarial framework GANcoder for training the auto-encoder on the textureless data. Traditional training approaches using an L1 norm tend to converge to the mean image on the low-textured images. In contrast, we use an adversarial discriminator to provided an additional loss function that is focused on distinguishing real images from the training dataset from the auto-encoder reconstruction. We collected a large GANPatches dataset of feature points from nearly textureless objects to train and evaluate our model and baselines. The dataset includes 16k pairs of image patches. We performed qualitative evaluation of our GANcoder and baselines for two tasks. Firstly, we compare the matching score of the our GANcoder and baselines. Secondly, we evaluate the accuracy of 3D reconstruction of low-textured objects using an SfM pipeline with stereo-matching provided by our GANcoder. The results of the evaluation are encouraging and demonstrate that our model achieves and surpasses the state of the art in the feature matching on low-textured objects.

The main aim of this research is to propose a novel technique for airborne SAR image registration using feature matching. The problem can be formulated as a model with three vital steps, including normalization, shape matrix estimation and geometrical invariant constraint algorithm. Starting from normalization, rotation angle between the two images can be obtained directly and independently. After rotation angle is corrected, this paper employs Shape Matrix (SM)estimation, one of the most classical feature descriptors to obtain a coarse translation matching. With the help of hierarchical strategy, an initial translation matching can be achieved rapidly and exactly. Finally, in this paper, cross ratio as geometrical invariant constraint is employed to detect and correct the mismatches due to SM estimation's deficiency. This is an optimization process with constraints based on projective invariance of cross-ratio of five coplanar feature points. Airborne SAR data acquired in eastern China in 2003 during the flood season is used for experiment, and the results show the method is effective and reliable.

ABSTRACTThis paper presents a new approach for geo-referencing historical maps using a polygon-based image registration technique. Since most historical maps lack long lasting point features to serve as control primitives for image registration, we explore the use of polygon features as control primitives that can be identified in both the map and the geo-referenced coordinate system based on matching the polygons’ shape context. A coordinate transformation model can be established using the matched vertices of the polygons, and the model coefficients are subsequently estimated using least-squares adjustment. The proposed method was tested on a digitized lithographic map of downtown Toronto, Ontario, Canada, created in 1857. The experimental work showed a good agreement for the image registration where the Dice similarity coefficient was 0.8 (i.e. 80% of overlap found between the two sets of polygons), regardless of using affine or polynomial model.