Abstract The basic principle of multi-view stereo (MVS) is to perform 3D reconstruction by extracting depth information from multiple views. Most current SOTA MVS networks are based on Vision Transformer, which usually means expensive computational complexity. To reduce complexity and improve map accuracy, we propose a network with Bidirectional Semantic Information (BSI-MVS). Firstly, design Multi-Level Spatial Pyramid module generate layers feature for multi-scale information. Then 2D...

A coarse-to-fine multi-view stereo network with Transformer (MVS-T) is proposed to solve the problems of sparse point clouds and low accuracy in reconstructing 3D scenes from low-resolution images. The uses a strategy estimate depth image progressively reconstruct cloud. First, pyramids features are constructed transfer semantic spatial information among at different scales. Then, module employed aggregate image's global context capture internal correlation feature map. Finally, inferred by...

In this paper, we present a method for digital image stabilization based on Fourier-Mellin transform and phase correlation. We acquire the rotating angle scaling factor firstly by correlation between images of reference observed images. After image, implement again to compute spacial translation. Because spectral periodicity, will be weaken maybe lead wrong result when is close 90°. So add coarse search before avoid situation. And use smooth window reduce noise in Furthermore, coordinate...

This paper presents a visual approach for distance measuring. Firstly, images are taken in different positions. Then the scaling parameter between these could be calculated with Fourier-Mellin transform. The transform is successfully utilized figuring out and rotation parameters images, meanwhile itself translation independent. So when rotated or translated, we can also get correspondingly accurate parameter. Finally, from optical center of camera to object figured pinhole model by two...

A digital image stabilization algorithm based on polar transform and circular block matching is utilized in this paper to stabilize the videos, taken from rolling airborne platform, which have rapid rotation arbitrary translation changes between adjacent frames. And used estimate global motion parameters, hierarchical search strategy reduce amount of computation. The experimental results show that proposed can achieve sub-pixel accuracy processing speed 20 frames per second for 360times264...

Infrared image simulation is challenging because it complex to model. To estimate the corresponding infrared directly from visible light image, we propose a three-level refined light-weight generative adversarial network with cascaded guidance (V2T-GAN), which can improve accuracy of image. V2T-GAN guided by cascading auxiliary tasks and information: first-level uses semantic segmentation as an task, focusing on structural information image; second-level grayscale inverted task supplement...

In this paper we present a novel log-polar transform based digital image stabilization algorithm to estimate large global transformations among consecutive frames. our proposed algorithm, multi- resolution techniques in spatial domain are introduced as an initial motion estimation module arbitrary rotations, translations and moderate scale changes. Then, gradient-based nonlinear least squares optimization is used achieve sub-pixel accuracy. The experimental results show that the can...

Unsupervised style transfer network can accomplish the task of infrared image simulation, and core its goal is to migrate images into visible while maintaining content images. However, existing methods have problems such as loss, edge blurring, poor stylization effect. In this paper, Multi-auxiliary Task Style Migration Network (MATST) proposed. The loss blurring are solved by adding coordinated attention generator generative adversarial network. simulation effect further improved semantic...

This paper presents a new digital image stabilization framework using an improved point matching algorithm. The in the is chosen lattice pattern, because can simulate different transform model. And motion estimated by lattice. algorithm compatible with models according to real experimental results. In situation of translation and rotation, coordinates be calculated stored advance reduce amount computation.

We present SA-MVSNet, a novel two-stage multi-view stereo network equipped with self-attention mechanism, which can improve the quality of low-resolution image 3D reconstruction. SA-MVSNet consists two stages, and lower resolution depth maps predicted in first stage provide priori information for second stage. To increase utilization information, pyramid scheme was used to fuse feature at different resolutions. Moreover, we introduce an improved module reconstruction accuracy by learning...

This paper presents an approach of global sparse matching algorithm based on Delaunay Triangle theory to obtain reliable result detected feature points between images. Our can solve the problem in the case images captured under a certain range scaling, rotation and translation, together with image affine distortion, addition noise, change illumination. Considering that it is hard obtain high percentage correct matched point pairs, we present this kind method improve accuracy exactness...

To estimate the 6-DoF camera pose from a single RGB image, we propose joint task learning (JTL) network called JTL-Net. JTL-Net is convolutional neutral with an asymmetric encoder–decoder structure consisting of shared bone as encoder, two independent decoders, and JTL regressor to output pose. Unlike most estimation networks, considers position orientation specific streams reduce interference caused by differences between positions angles. The regressor, attention-guided architecture,...

A novel robust point matching method is presented in this paper, which can find corresponding pairs effectively via 2D or 3D points' relative positions. There the positions of sets come from same moving rigid object are invariable. To describe position, a vector structure constructed by connecting one to rest. If two points different pair, their should be similar, vectors will have approximate lengths and mutual angles. Our based on kind similarity. Firstly we apply an algorithm Hausdorff...