image repair is to repair partially damaged images. At present, when repairing the images with arbitrary missing shape and the images with large defect area, the current methods have some problems, such as fuzzy image...
详细信息
image repair is to repair partially damaged images. At present, when repairing the images with arbitrary missing shape and the images with large defect area, the current methods have some problems, such as fuzzy image repair and differences in the repaired joints. Therefore, this paper proposes an image repair model of MRS-Net(Multiscale Residual Squeeze-and-congestion Networks) by using the generated countermeasure network to repair the images with arbitrary missing shape and large defect area. The generator adds a residual attention module in the connection layer between the encoder and the decoder to improve the repair ability of the model, and generates a network through multi-scale joint feedback against loss, reconstruction loss, perception loss, style loss and total variation loss, so as to ensure the visual consistency between the repair boundary and the surrounding real image. At the same time, the binary cross entropy loss feedback discriminant network is used. The proposed model is trained and tested on the dataset CelebA and Oxford buildings. Experiments show that the proposed model can effectively extract the missing information, Meanwhile, the repair results have natural transition boundaries and clear details. MRS-Net can improve SSIM(structural similarity) index by about 2% - 5% and PSNR(peak signal-to-noise ratio)index by about 1-3. The FID(Frechet Inception Distance score) straight index is reduced by 2-8, and the evaluation indexes are *** the missing area accounts for 5% - 10%, 11% - 20%, 21% - 30%, 31% - 40% and 41% - 50%,MRS-Net has better image restoration effect, which can reflect that MRS-Net has better *** proposed image restoration method can repair faces, buildings and other scenes. At the same time, it can repair defects in different shapes and areas.
Focused on the issue that the object structure discontinuity and poor texture detail occurred in imageinpainting method, the image inpainting algorithm based on self-adaptive group structure has proposed in this pape...
详细信息
Focused on the issue that the object structure discontinuity and poor texture detail occurred in imageinpainting method, the image inpainting algorithm based on self-adaptive group structure has proposed in this paper. The conception of self-adaptive group structure is different from traditional image patching operation and fixed group structure, which refers to the fact that a patch on the structure has fewer similar patches than the one within the textured region. A self-adaptive dictionary as well as the sparse representation model was established in the domain of self-adaptive group. Finally, the target cost function was solved by Split Bregman Iterational operation. The experimental results on target removing with Criminisi's algorithm, GSR's algorithm and SALSA's algorithm in image pixels losting of imageinpainting had shown that the proposed algorithm has better performance than other algorithms.
image compression is an intensively studied subject in computer vision. The deep generative model, especially generative adversarial networks (GANs), is a popular new direction for this subject. In this study, the aut...
详细信息
image compression is an intensively studied subject in computer vision. The deep generative model, especially generative adversarial networks (GANs), is a popular new direction for this subject. In this study, the authors propose a new compression method based on a generative model and focus on its application by GANs. The decoder in the proposed method is modified from the GAN generator model, which can produce visually real-like synthetic images. It is one of the two models in GANs, which is trained through a two-players' contest game. The encoder is an optimisation algorithm called backpropagation-to-the-input, which derives from an image inpainting algorithm based on generative models. In the proposed method, the authors turn the encoding process into an optimisation task to search for optimal encoded representations. Compared with traditional methods, the proposed method can compress images from certain domains into extremely small and shape-fixed encoded space but still retain better visual representations. It is easy and convenient to apply without any retraining or additional modification to the generative models.
In this paper, we introduce the new approach to enhance the reliability of detection of objects in a driving environment (e.g. pedestrian and vehicle). We present the method of filtering out false positive detections ...
详细信息
ISBN:
(纸本)9781457721977
In this paper, we introduce the new approach to enhance the reliability of detection of objects in a driving environment (e.g. pedestrian and vehicle). We present the method of filtering out false positive detections while maintaining true positive detections. Our approach considers that if we remove a certain region from an image taken from a vehicle in a driving environment, the inpaintingalgorithm is able to restore the removed region based on its surroundings when it does not include objects. Previous inpaintingalgorithms were used for restoration of damaged paintings, and we expand its usage to confirm whether the detection result includes the real object or not. Furthermore, we introduce a simple but effective speed-up method for the sliding window using simple edge features of objects. Experimental results confirm that our approach is able to improve the accuracies of various pedestrian and vehicle detectors. We show the improved accuracy of pedestrian and vehicle detection in a driving environment with various detectors.
暂无评论