This paper proposes a Bayesian theory of hierarchical cortical computation based on mathematical and computational ideas from computer vision and pattern theory, as well as recent neurophysiological evidence. The theory suggests that early visual neurons in areas V1 and V2 are not static filters but are part of a complex, interactive system that integrates top-down contextual priors and bottom-up observations through recurrent feedforward/feedback loops. This framework allows for concurrent probabilistic inference along the visual hierarchy. The authors suggest that algorithms like particle filtering and Bayesian belief propagation could model these cortical computations. They review recent neurophysiological evidence supporting this theory, indicating that early visual cortex activity is tightly coupled with the rest of the visual system. The paper discusses the hierarchical Bayesian inference framework, which provides an alternative perspective for understanding recent neurophysiological findings. It emphasizes that inference is more general than competition and that feedback should be viewed as mechanisms for biasing inference and computations along the visual hierarchy, not just as attentional selection or biased competition. The framework is applied to the visual cortex, where V1 is considered a high-resolution buffer for integrating information into a coherent percept. The paper also discusses the role of V1 in processing local visual features and how feedback from higher areas can refine these features. The authors argue that the hierarchical Bayesian framework can explain the interaction among multiple cortical areas, such as V1, V2, V4, and the inferotemporal cortex (IT). They propose that the recurrent feedback connections among these areas allow for the constraint of one another's computations, enabling concurrent processing across multiple levels. The paper also discusses the role of particle filtering in visual processing, noting its success in tracking moving objects in cluttered environments and its application in robotics for mapping and localization. The paper presents experimental evidence supporting the hierarchical Bayesian framework, including studies on the timing of visual processing, the scale of analysis, and the interactive hierarchy of visual areas. These experiments suggest that feedback is essential for processing ambiguous scenes and that higher-level areas can influence lower-level areas through top-down signals. The authors also discuss the role of V1 in contour completion and shape from shading, showing that V1 neurons can respond to illusory contours, indicating the importance of feedback in early visual processing. The findings support the idea that the visual system uses a generative model to interpret scenes, with feedback playing a crucial role in this process.This paper proposes a Bayesian theory of hierarchical cortical computation based on mathematical and computational ideas from computer vision and pattern theory, as well as recent neurophysiological evidence. The theory suggests that early visual neurons in areas V1 and V2 are not static filters but are part of a complex, interactive system that integrates top-down contextual priors and bottom-up observations through recurrent feedforward/feedback loops. This framework allows for concurrent probabilistic inference along the visual hierarchy. The authors suggest that algorithms like particle filtering and Bayesian belief propagation could model these cortical computations. They review recent neurophysiological evidence supporting this theory, indicating that early visual cortex activity is tightly coupled with the rest of the visual system. The paper discusses the hierarchical Bayesian inference framework, which provides an alternative perspective for understanding recent neurophysiological findings. It emphasizes that inference is more general than competition and that feedback should be viewed as mechanisms for biasing inference and computations along the visual hierarchy, not just as attentional selection or biased competition. The framework is applied to the visual cortex, where V1 is considered a high-resolution buffer for integrating information into a coherent percept. The paper also discusses the role of V1 in processing local visual features and how feedback from higher areas can refine these features. The authors argue that the hierarchical Bayesian framework can explain the interaction among multiple cortical areas, such as V1, V2, V4, and the inferotemporal cortex (IT). They propose that the recurrent feedback connections among these areas allow for the constraint of one another's computations, enabling concurrent processing across multiple levels. The paper also discusses the role of particle filtering in visual processing, noting its success in tracking moving objects in cluttered environments and its application in robotics for mapping and localization. The paper presents experimental evidence supporting the hierarchical Bayesian framework, including studies on the timing of visual processing, the scale of analysis, and the interactive hierarchy of visual areas. These experiments suggest that feedback is essential for processing ambiguous scenes and that higher-level areas can influence lower-level areas through top-down signals. The authors also discuss the role of V1 in contour completion and shape from shading, showing that V1 neurons can respond to illusory contours, indicating the importance of feedback in early visual processing. The findings support the idea that the visual system uses a generative model to interpret scenes, with feedback playing a crucial role in this process.