44
J.Jensen
classification, or complex spatial GIS modeling. The output from the intensive
mainframe processing can be passed to a workstation or personal computer for
subsequent less expensive processing if desired (Davis, 1993). Mainframe computer systems are expensive to purchase and maintain.
3.1.2 Number of Aualysts ou a System and Mode of Operatiou
The ideal digital image processing environment is when a single user sits in front
of single workstation. Unfortunately, this is not always possible due to cost constraints. The sophisticated workstation lab shown in Fig. 3.1 might be ideal for
research, but ineffective for education or short course instruction where many
analysts (e.g. > 20) must be served.
It is well known that the best scientific visualization environment takes place
when the digital image processing system uses an interactive graphical-userinterface (GUI) (Mazlish, 1993; Miller and DeCampo, 1994). Two effective
graphical user interfaces include ERDAS Imagine's intuitive point and click icons
(Fig. 3.2a, b) and ENVI's hyperspectral data analysis interface (Fig. 3.3). Late
night non-interactive batch processing is of value for time consuming processes
(e.g. resampling during image rectification) and helps to free-up lab workstations
during peak demand.
3.1.3 Serial versus Parallel Image Processing, Arithmetic Coprocessor, and
Random Access Memory (RAM)
Some computers have multiple CPUs that operate concurrently. Parallel processing software allocates the CPUs to perform efficient digital image processing
(Faust et aI;, 1991). For example, consider a 512 node (CPU) pm'allel computer. If
a remote sensing dataset consisted of 512 pixels (columns) in a line, each of the
512 CPUs could be programmed to process an individual pixel, speeding up the
processing of a single line of data by 512 times. If 512 bands of hyperspectral data
were available, each processor could be allocated to a single band to perform independent processing. Many vendors are developing digital image processing
code that takes advantage of parallel architecture.
An arithmetic coprocessor is a special mathematical circuit that performs highspeed floating point operations while working in harmony with the CPU. Most
sophisticated image processing software often will not function without a math
coprocessor. If substantial resources are available, then an array processor is ideal.
It consists of a bank of memory dedicated to performing simultaneous computations on elements of an array (matrix) of data in n dimensions (Freedman, 1995).
Remotely sensed data are collected and stored as arrays of numbers so array processors are especially well suited to image enhancement and analysis operations.
Random access memory (RAM) is the computer's primary temporary workspace. RAM chips require power to maintain their content. Therefore, all information stored in RAM must be saved to a hard disk (or other media) before turning
off the computer. The computer should contain sufficient RAM for the operating
system, image processing applications software, and any remote sensor data that
must be held in memory while calculations are performed. Computers with 64-bit
J.Jensen
classification, or complex spatial GIS modeling. The output from the intensive
mainframe processing can be passed to a workstation or personal computer for
subsequent less expensive processing if desired (Davis, 1993). Mainframe computer systems are expensive to purchase and maintain.
3.1.2 Number of Aualysts ou a System and Mode of Operatiou
The ideal digital image processing environment is when a single user sits in front
of single workstation. Unfortunately, this is not always possible due to cost constraints. The sophisticated workstation lab shown in Fig. 3.1 might be ideal for
research, but ineffective for education or short course instruction where many
analysts (e.g. > 20) must be served.
It is well known that the best scientific visualization environment takes place
when the digital image processing system uses an interactive graphical-userinterface (GUI) (Mazlish, 1993; Miller and DeCampo, 1994). Two effective
graphical user interfaces include ERDAS Imagine's intuitive point and click icons
(Fig. 3.2a, b) and ENVI's hyperspectral data analysis interface (Fig. 3.3). Late
night non-interactive batch processing is of value for time consuming processes
(e.g. resampling during image rectification) and helps to free-up lab workstations
during peak demand.
3.1.3 Serial versus Parallel Image Processing, Arithmetic Coprocessor, and
Random Access Memory (RAM)
Some computers have multiple CPUs that operate concurrently. Parallel processing software allocates the CPUs to perform efficient digital image processing
(Faust et aI;, 1991). For example, consider a 512 node (CPU) pm'allel computer. If
a remote sensing dataset consisted of 512 pixels (columns) in a line, each of the
512 CPUs could be programmed to process an individual pixel, speeding up the
processing of a single line of data by 512 times. If 512 bands of hyperspectral data
were available, each processor could be allocated to a single band to perform independent processing. Many vendors are developing digital image processing
code that takes advantage of parallel architecture.
An arithmetic coprocessor is a special mathematical circuit that performs highspeed floating point operations while working in harmony with the CPU. Most
sophisticated image processing software often will not function without a math
coprocessor. If substantial resources are available, then an array processor is ideal.
It consists of a bank of memory dedicated to performing simultaneous computations on elements of an array (matrix) of data in n dimensions (Freedman, 1995).
Remotely sensed data are collected and stored as arrays of numbers so array processors are especially well suited to image enhancement and analysis operations.
Random access memory (RAM) is the computer's primary temporary workspace. RAM chips require power to maintain their content. Therefore, all information stored in RAM must be saved to a hard disk (or other media) before turning
off the computer. The computer should contain sufficient RAM for the operating
system, image processing applications software, and any remote sensor data that
must be held in memory while calculations are performed. Computers with 64-bit