Main Stages of Visual Information Processing

3. Two Main Stages of Visual Information Processing

3.1 Bottom-up Processing

Initially, a base representation is cxreated.  This representation provides a local description of the visual input and includes information of objects such as depth, orientation, color, motion etc. with respect to the viewer (e.g. this representation may change if the viewer's position / orientation changes).  This representation is dependant on the visual input only - no higher level processing is performed to the input, does not rely on memory or any cognitive processes.  Furthermore, this representation remains fixed and is not modified.  As a result, a change in the viewer's posaition / orientation would result in the creation of a new base representation.

3.2 Top-down Processing

Visual routines are applied to the base representation in order to obtain properties and relations between the objects in the base representation.  This step relies on more than just the visual input and may use memory and other cognitive processes and is applied to a portion of the base representation only.

3.2.1 Incremental Representations

While the visual routines are performing computations, other "middle" representations are created and modified during the processing in order to maintain the results obtained after the visual routines perform their opewrations.  These representations depend on the particular visual routines applied and the same input on different occasions mauy lead to a different representation whereas the same input will always lead to the same base representation.  Other routines may then take advantage of the incremental represenations allowing them to operate more efficiently and faster as a result.  As an example of an incremental representation, consider determining whether some target is inside some polygon by using for example area coloring.  If it is determined that the target is inside the polygon, then it may have been determined that the polygon is closed.  The fact that the polygon is closed will be stored in the incremental operation and any other routine which may need to know if the polygon is closed, will not have to apply any elemental operations to determine this.

Summary of the Role of Visual Routines in the Processing of Visual Information
 

Figure 7.  Visual Routine Processor is responsible for putting together (e.g. combining the elemental operations) and executing the visual routines.  The processor also acts as a "middle-man" between the base and incremental representations and the higher level components such as memory.  Communications between them is done via the routine processor.

3.2.2 Parallel Processing of Visual Information

Is the visual system capable of performing operations in parallel?  It appears as though it does and the aurthor agrees.  Furthermore, the author defines three different types of parrallel processing which the visual system is capable of.
  1. Spatial:  the same operation applied simultaneously at different locations.
  2. Functional: different operations applied simultaneously the same location.
  3. Temporal: the "simultaneous application of different processing stages to different inputs" (also known in the field of computer architecture as pipelining).
Visual routines are capable of operating in all three parallelisms, however, there appears to be a limited use of spatial parallelism!