计算机视听觉－人工智能的梦.ppt

计算机视听觉－人工智能之梦Computer Seeing and Hearing-A Dream of AI 张钹 清华大学信息科学与技术学院 清华大学计算机科学与技术系 清华信息科学与技术国家实验室 智能技术与系统国家重点实验室 The Characteristic of Auditory Information (Data) The Characteristics of Visual Information (Data) The Sparseness of Auditory Signal The Sparseness of Visual Signal 一个不适定问题An Ill-posed Problem Image Segmentation vs. Recognition Speech Segmentation vs. Recognition 技术上的困难(Technology) The Relation Between Activation Patterns and Early Stages of Sound Processing Multi-layer Neural Network with feedback connections Representation Experimental Results Computer Robustly Extractable Features Statistical Method Representation at Different Granularities Pixel-based Representation-the finest representation Global Features -the coarsest representation Coarse vs. Fine Representation Representation with Middle Grain-Size Local (Spatial) Feature Vector Representation Region-adaptive Grid Partition Hierarchical（粒度）结构 Experiments TRECVID Data Coarse vs. Fine Granulation Multi-granulation Multi-granulation (2) Multi-Granular Multi-modal PMSRA The Comparison between Uni-modal and Multi-granular, modal 声波、声谱图（Spectrograms） 语音信息 不同粒度的语音特征 语音的时间结构 (Temporal Structure) Image Region Annotation -horse, sky, mountain, grass, tree Region-adaptive Grid Partition (2) Experiments Image Region Annotation Image Region Annotation Spatial Structural Representation Different Models Label Configuration Comparison Among Different Models Experimental Results The Spatial Relation Among Region Labels 数据驱动法（Data-driven） The Bless of Dimensionality Anti-Occluded Two Basic Problems 基于人类认知的图像处理 Cognition (Perception) Space Object Recognitionwith sparse, localizedfeatures Computational Model Experimental Results 人脑听觉皮层的试验研究 Sparse Representation Sample Space (Data Space) Extended Yale B 2414 frontal-face with different lighting 38 individuals 192?168 image J. Wright, et al. Robust face recognition via sparse representation, IEEE PAMI 08 Anti-Noise 30% 50% 70%