对象标识，即一个对象被看作是相同的对象，尽管在其他属性（例如，照明，位置，闭塞）的实质性变化，是视觉系统的主要挑战之一。和在复杂环境中的有意义的物体检测识别是一个关键技能的基础广泛的行为，从医学影像诊断识别猎物和捕食者，从认识到食用食物在人群中寻找熟悉的面孔。在人类和灵长类动物的视觉能力，这些过程都是毫不费力地，快速，自动地。然而，目标识别的计算挑战是远离琐碎。特别是，相干有意义的对象的识别涉及到集成的组件在不同层次的视觉复杂度，从初级视觉区域的局部轮廓到复杂的对象在更高层次的模块。这些表示是高度宽容的环境变化，如观点，大小，位置，背景杂波等变化。
许多研究现在提供融合的证据，涉及在物体识别的神经基板主要位于腹侧视觉流。潜在的机制似乎是由一个组织的处理阶段，在粗糙的层次顺序。信息从视网膜V1投射，从V1到高级皮层区结束了AIT（前颞皮层）猴被认为是同源的人类地区LOC（侧枕叶）和fusifurmgyrus。这些区域，不同于较低的区域，编码对象标识。例如，它已被证明，在区域特的它（下颞）皮层的猴子的大脑的细胞选择性地响应各种复杂的对象特征。此外，该神经元相似的功能都聚集在柱区域伸出垂直于皮层表面（tanifuji等人。2001）。此外，这些神经元代表复杂的视觉对象的使用人口稀疏编码（Tanaka等人。1991田中1999）。另一方面，一些功能磁共振成像研究已经显示出类似的对象在人体腹侧通路的分布表示。例如，研究对比的面孔，房子和椅子已经确定双侧腹颞侧和腹侧枕叶皮质和颞上沟的差别激活地图（Ishai等。（1999），Ishai等人（2000）。他们提出，对象表示在腹侧视觉通路是由一个关于形态信息的分布式表示这样做并不是由小高选择性的皮质补丁。图1勾画了一个类的对象识别模型，从视网膜输入结束在分类和识别，并显示了一个类的对象识别模型的示意图。
莫纳什大学代写作业: 高层适应、目标识别
Object identification, whereby an object is seen as the same object despite substantial changes in other attributes (e.g., illumination, location, occlusion), is one of the major challenges for the visual system. The detection and recognition of meaningful objects in complex environments is a crucial skill that underlies a wide range of behaviors, from diagnosing tumors on medical images to identifying preys and predators, and from recognizing edible foods to finding familiar faces in a crowd. In humans and primates with visual abilities, these processes operate effortlessly, quickly and automatically. However, the computational challenges of object recognition are far from trivial. In particular, the recognition of coherent meaningful objects involves integration of components at different levels of visual complexity, starting from local contours at primary visual areas to complex object at higher level modules. These representations are highly tolerant to environmental variations such as changes in viewpoint, size, position, background clutter, etc.
Many studies now provide converging evidence that the neural substrates involved in object recognition are mainly located in the ventral visual stream. The underlying mechanism appears to consist of a number of processing stages that are organized in rough hierarchical order. The information is projected from retina to V1, and from V1 to higher cortical areas ending up at AIT (Anterior Inferior Temporal Cortex) in monkey which is considered homologous to human area LOC (Lateral Occipital Complex) and the fusifurmgyrus. These areas, unlike the lower ones, encode object identity. For example, it has been shown that cells in area TE of the IT (inferior temporal) cortex of the monkey brains respond selectively to various complex object-features. Moreover, the neurons that respond to similar features are clustered in columnar region stretched out vertically to the cortical surface (Tanifuji et al. 2001). Furthermore, these neurons represent the complex visual objects using a sparse population code (Tanaka et al. 1991; Tanaka 1999). On the other hand, several fMRI studies have shown a similar distributed representation of objects in human ventral pathway. For example,studies contrasting faces, houses and chairs have identified differential activation maps bilaterally in ventral temporal and ventral occipital cortices and superior temporal sulcus (Ishai et al. (1999), Ishai et al.(2000). They propose that bject representation in the ventral visual pathway is done by a distributed representation of information about the form and not by small highly selective cortical patches. Figure 1 sketches a class of object recognition models starting from the retinal input ending up in categorization and identification and shows a sketch of a class of object recognition models.
如果您在墨尔本读书，需要代写服务，可以联系我们的客服，或查看 墨尔本论文代写
