The proceedings contain 155 papers. The topics discussed include: two fundamental challenges in perceptual coding and image restoration;robust recognition-by-parts using transduction and boosting with applications to ...
ISBN:
(纸本)9789612480295
The proceedings contain 155 papers. The topics discussed include: two fundamental challenges in perceptual coding and image restoration;robust recognition-by-parts using transduction and boosting with applications to biometrics;overview of multi-view video coding;data protection techniques, cryptographic protocols and PKI systems in modern computer networks;analysis of fused ophthalmologic image data;threshold estimation for wavelet domain filtering of signal-dependent noise;partial realization of a generalized transfer function;analysis of electrocardiograms using the convolution kernel compensation approach;using fractal dimension as texture discriminator for content base image retrieval;image interpolation method based on wavelets;face feature detection for 3D model of talking head with speech synthesis;chromatic enhancement technique for JPEG image;and a formal approach to hypervideo design.
作者：
Chen, ZhaoguoCollege of Arts
Shandong Agricultural Engineering University Shandong Province Jinan250103 China
To fully harness the capabilities of computer graphics and imageprocessing technologies and elevate the quality of visual communication design, this paper presents a comprehensive suite of innovative methodologies. F...
详细信息
Zero-shot learning (ZSL) directs the challenge of classifying unseen test images without explicit training on those samples. ZSL can identify and classify unlabeled images available in abundance by learning from visua...
详细信息
ISBN:
(纸本)9783031734762;9783031734779
Zero-shot learning (ZSL) directs the challenge of classifying unseen test images without explicit training on those samples. ZSL can identify and classify unlabeled images available in abundance by learning from visual and semantic embedding vectors (feature vectors). Information-enriched visual features extracted from images play a crucial role in ZSL. This paper proposes a hybrid feature approach that integrates low-level (LL), and high-level (HL) features extracted from images. Gray Level Co-occurrence Matrix (GLCM) and Gabor features are employed to obtain LL texture features, while HL features are derived from the ResNet-50 model, renowned for capturing complex hierarchical representations. These hybrid visual features are then mapped with semantic features using linear mapping, where the semantic features are embedding vectors of labels generated by the fastText model. Experiments on the AWA2 and SUN datasets are conducted in a bid to evaluate the proposed approach's effectiveness. The hybrid feature approach has demonstrated enhanced quality in zero-shot image classification, effectively classifying images that the model has not seen during training.
Text-to-image generation is a cutting-edge technology that enables computers to generate images from textual descriptions. While this technology has been extensively researched and applied to English language text, ap...
详细信息
ISBN:
(纸本)9783031804373;9783031804380
Text-to-image generation is a cutting-edge technology that enables computers to generate images from textual descriptions. While this technology has been extensively researched and applied to English language text, applying it to Arabic language text is still in its early stages. Additionally, the Arabic language is challenging due to its right-to-left writing system and extensive vocabulary of 1.3 million words. In this paper, we explore text-to-image generation for generating images from Arabic language text descriptions. Firstly, we fine-tune a transformer-based model pre-trained on the Arabic text to transform the text information into affine transformation within the DF-GAN generator. Secondly, we present a text transformer that combines LSTM layers to address the limitation of unrecognized words. Thirdly, a mask predictor is trained into the generator using a weakly supervised method and incorporated into the affine transformation for a more effective integration of image and text features. In addition, we add the DAMSM loss function as a regularization to the loss function to achieve convergences and stability in the training phase. The experiment on two challenging datasets CUB and Oxford-flower shows that our architectures can accurately generate high-quality images faithfully representing the Arabic textual descriptions. We believe the scaling of this task could have critical applications in fields such as Arabic visual learning, e-commerce, advertising, and entertainment.
The process that produces written descriptions that effectively represent the meaning and context of an image is known as image captioning. To integrate visual and textual data, it needs to blend computer vision and n...
详细信息
The proceedings contain 39 papers. The special focus in this conference is on imageprocessing and communications. The topics include: Two-Dimensional hidden Markov models in road signs recognition;evaluating the mutu...
ISBN:
(纸本)9783319106618
The proceedings contain 39 papers. The special focus in this conference is on imageprocessing and communications. The topics include: Two-Dimensional hidden Markov models in road signs recognition;evaluating the mutual position of objects on the visual scene using morphological processing and reasoning;clustering-based retrieval of similar outfits based on clothes visual characteristics;improving shape retrieval and classification rates through low-dimensional features fusion;accelerating the 3D random walker image segmentation algorithm by image graph reduction and GPU computing;computed tomography images denoising with Markov random field model parametrized by Prewitt mask;Gaussian mixture model based non-local means technique for mixed noise suppression in color images;robust image retrieval based on mixture modeling of weighted Spatio-color information;noise reduction in ultrasound images based on the concept of local neighbourhood exploration;Viterbi algorithm for noise line following robots;hybrid shape descriptors for an improved weld defect retrieval in radiographic testing;on the usefulness of combined metrics for 3D image quality assessment;imageprocessing with process migration;comparison of assessment regularity methods dedicated to isotropic cells structures analysis;a texture-based energy for active contour image segmentation;object localization and detection using variance filter;the impact of the image feature detector and descriptor choice on visual SLAM accuracy and shape versus texture.
The proceedings contain 131 papers. The topics discussed include: multimedia in Croatian digital video broadcasting;signal and image data processing in ultrasonic imaging;progressive image coding using regional color ...
ISBN:
(纸本)9531840547
The proceedings contain 131 papers. The topics discussed include: multimedia in Croatian digital video broadcasting;signal and image data processing in ultrasonic imaging;progressive image coding using regional color correlation;edge-preserving regularization of disparity and motion fields;an edge-preserving high-resolution image reconstruction;a new method for image coding in computer vision;efficient post-processing for block-based compressed video;detecting lesions in a mammogram;a methodology for evaluating the operational effectiveness of facial recognition systems;gallery image effects on facial recognition systems;on edge detection in MRI using the wavelet transform and unsupervised neural networks;improved illumination independent moving object detection in real world video sequences;and virtual image : keyframe or visual icon?.
The proceedings contain 131 papers. The topics discussed include: multimedia in Croatian digital video broadcasting;signal and image data processing in ultrasonic imaging;progressive image coding using regional color ...
ISBN:
(纸本)9531840547
The proceedings contain 131 papers. The topics discussed include: multimedia in Croatian digital video broadcasting;signal and image data processing in ultrasonic imaging;progressive image coding using regional color correlation;edge-preserving regularization of disparity and motion fields;an edge-preserving high-resolution image reconstruction;a new method for image coding in computer vision;efficient post-processing for block-based compressed video;detecting lesions in a mammogram;a methodology for evaluating the operational effectiveness of facial recognition systems;gallery image effects on facial recognition systems;on edge detection in MRI using the wavelet transform and unsupervised neural networks;improved illumination independent moving object detection in real world video sequences;and virtual image : keyframe or visual icon?.
暂无评论