RAPID DEVELOPMENT OF RETINEX ALGORITHM ON TI

C6000-BASED DIGITAL SIGNAL PROCESSOR

Juan Zapata and Ram´on Ruiz

Departmento de Electr´onica, Tecnolog´ıa de Computadoras y Proyectos

Universidad Polit´ecnica de Cartagena,Muralla del Mar, s/n, Cartagena, Spain

Keywords:

Retinex Algorithm, Real-Time Signal Processing, Real-Time Workshop, MATLAB, Simulink.

Abstract:

The Retinex is an image enhancement algorithm that improves the brightness, contrast and sharpness of an

image. This work discusses an easy and rapid DSP implementation of the Retinex algorithm on a hard-

ware/software platform which integrates MATLAB/Simulink, Texas Instruments (TI) eXpressDSP Tools and

C6000 digital signal processing (DSP) target. This platform automates rapid prototyping on C6000 hard-

ware targets because lets use Simulink to model the Retinex algorithm from blocks in the Signal Processing

Blockset, and then use Real-Time Workshop to generate C code targeted to the TI DSP board by mean Code

Composer Studio (CCS IDE). The build process downloads the targeted machine code to the selected hard-

ware and runs the executable on the digital signal processor. After downloading the code to the board, our

Retinex application runs automatically on our target. It performs a non-linear spatial/spectral transform that

synthesizes strong local contrast enhancement. The library real time data exchange (RTDX) instrumentation

that contains RTDX input and output blocks let transfer image to and from memory on any C6000-based

target.

1 INTRODUCTION

With the rapid evolution in semiconductor technol-

ogy in the past several years, digital signal processing

systems have a lower overall cost compared to ana-

log systems. DSP applications can be developed, an-

alyzed, and simulated using software tools (Chacon

and Valenzuela, 2004)(Assis de Melo and La Neve,

2004)(Gan and Kuo, 2006).

Due to the fast-paced nature of the digital signal

processing applications and to the limited life span of

new products, the time to market (TTM) is a very im-

portant ﬁgure of merit that is often overlooked. The

rapid realization of an implementation from its con-

cept to a product is of utmost importance. The sci-

entiﬁc community in general, and the signal process-

ing community in particular, have developed a num-

ber of methods for the speciﬁcation of higher level

algorithmic concepts and ideas. Two equivalents al-

ternatives are graphical methods and language-based

methods. Graphical method includes block diagrams,

state diagrams and schemes for the design of vir-

tual prototypes and language-based method includes

hardware description languages (HDLs). Simulink

uses graphical block diagrams to create models for

real-time implementation of applications and then use

Real-Time Workshop to generate C code targeted to

the TI DSP board by mean Code Composer Studio

(CCS IDE). The Retinex is an image enhancement al-

gorithm based on human perception which provides

color constancy and dynamic range compression. The

algorithm is inspired by the work of Land’s model

for human vision. This algorithm was initially tar-

geted to process multi-spectral satellite imagery but

has found applicability in very diverse areas such

as medical radiography, medical ultrasound imagery,

aviation safety, and others. This paper presents the

rapid prototyping and implementation of Retinex im-

age enhancement algorithm on a platform based on

a TI C6000 DSP target and Simulink/MATLAB/CCS

development platform.

149

Zapata J. and Ruiz R. (2007).

RAPID DEVELOPMENT OF RETINEX ALGORITHM ON TI C6000-BASED DIGITAL SIGNAL PROCESSOR.

In Proceedings of the Second International Conference on Computer Vision Theory and Applications - IFP/IA, pages 149-152

 SciTePress

2 SCHEME OVERVIEW

The key aspect of rapid prototyping is automated code

generation. Under our scheme, the algorithm for a

given application is initially described with signal-

ﬂow block diagrams with Simulink. Simulink is

a platform for multidomain simulation and Model-

Based Design for dynamic systems. It provides an

interactive graphical environment and a customizable

set of block libraries, and can be extended for spe-

cialized applications. Models built in Simulink can

be conﬁgured and made ready for code generation.

Using Real-Time Workshop, C code can be gener-

ated from the model for real-time simulation, rapid

prototyping, or embedded system deployment. Fig-

ure 1 shows the general scheme for rapid prototyping

based on MATLAB/Simulink and Texas Instruments

eXpressDSP tools.

Link for Code Composer Studio

MATLAB

Embedded Target

For TI C6000

RTDX

Target Language Compiler

Code Composer Estudio IDE

DSP C6000

C Code Model

Block Model

RTDX Comm.

Real-Time Workshop

Executable Model

Simulink

Figure 1: Development scheme of MathWorks and Texas

Instruments eXpressDSP tools.

Real-Time Workshop generates and executes

stand-alone C code for developing and testing al-

gorithms modeled in Simulink. The resulting code

can be used for many real-time and non-real-time

applications, including simulation acceleration, rapid

prototyping, and hardware-in-the-loop testing. Real-

Time Workshop uses target template ﬁles to trans-

late Simulink models into ANSI/ISO C code (Gan,

2002) (Chassaing, 2002). The target templates spec-

ify the environment on which this generated code

will run. Own custom targets can be develop or

use the ready-to-run conﬁgurations and third-party

targets supported by Real-Time Workshop. For-

tunelly, the Embedded Target for TI TMS320C6000

DSP (User Guide 3, 2006) consists of TI C6000 target

(C6000lib blockset) that automates rapid prototyping

on a C6000 hardware targets.

Embedded Target for TI TMS320C6000 DSP in-

tegrates Simulink and MATLAB with Texas Instru-

ments eXpressDSP tools. TI development tools

pick the C code generated with Real-Time Work-

shop (User Guide 6, 2006b) for a customized hard-

ware target supported for Embedded Target for TI

TMS320C6000 DSP and build an executable ﬁle for

this target-speciﬁc processor. Additionally, one of

Real-Time Workshop build options builds a Code

Composer Studio project from the C code generated

and, therefore, all features provided by Code Com-

poser Studio work to help develop the algorithm or

application.

Once target-speciﬁc executable is downloaded to

the hardware and run it, the code runs wholly on the

target and the running process can be accessed only

from Code Composer Studio or from MATLAB with

two powerful tools: Link for Code Composer Stu-

dio (User Guide 6, 2006a) and Real-Time Data Ex-

change (RTDX).

Link for Code Composer Studio lets use MAT-

LAB functions to communicate with Code Composer

Studio and with the information stored in memory and

registers on a target. With the links, information can

be transfered to and from Code Composer Studio and

with the embedded objects, information about data

and functions stored on the signal processor can be

retrieved. Within the collection of hardware that Link

for Code Composer Studio supports, some features of

the link cannot be applied.

3 SYSTEM CONFIGURATION

In our scheme, the building process is initiated from

Simulink, with a model or algorithm. Next, the Target

language Compiler from Real-Time Workshop builds

a program automatically for real-time application in

C6000 environment. Using the make utility, Real-

Time Workshop controls how it compiles and links

the generated source code. Data can be sent, or re-

ceived to the application through the RTDX channels.

3.1 Real-Time Workshop Conﬁguration

Real-Time Workshop analyzes the block diagram and

compiles an intermediate hierarchical representation

in a ﬁle called

model.rtw

. The Target Language

Compiler reads

model.rtw

, translates it to C code.

Real-Time Workshop constructs a makeﬁle from the

appropriate target makeﬁle template. The make utility

reads the makeﬁle to compile source code, link object

ﬁles and libraries, and generate an executable image,

called model (UNIX) or model.exe (Windows). Fig-

ure 2 illustrates the complete process. The box la-

beled ”Automated build process” highlights portions

of the process that Real-Time Workshop executes.

After generating the code, Real-Time Workshop

generates a customized makeﬁle,

model.mk

. The

generated makeﬁle instructs the make system utility

to compile and link source code generated from the

model, as well as any required harness program, li-

braries, or user-provided modules. Real-Time Work-

shop creates

model.mk

from a system template ﬁle,

system.tmf

and where system stands for the selected

target name. The system template makeﬁle is de-

signed for a speciﬁc target environment. Exits the

option of modifying the template makeﬁle to specify

compilers, compiler options, and additional informa-

tion used during the creation of the executable.

Real-Time Workshop creates the

model.mk

ﬁle

by copying the contents of

system.tmf

and expand-

ing lexical tokens (symbolic names) that describe a

model’s conﬁguration. Real-Time Workshop pro-

vides many system template makeﬁles, conﬁgured

for speciﬁc target environments and development sys-

tems.

4 DEVELOPMENT OF RETINEX

ALGORITHM

4.1 Retinex

The general form of the center/surround retinex is

similar to the difference of Gaussian (DOG) function

widely used in natural vision science to model both

the perceptive and receptive processes. Expressed

mathematically, this take the form

(x,y) = logI

(x,y) − log[F(x,y) ∗ I

(x,y)] (1)

where I

(x,y) is the image distribution in the ith color

spectral band, “*” denotes the convolution operation,

F(x,y) is the surround or kernel function, and R

(x,y)

is the associated retinex output.

F(x,y) is a Gaussian Filter deﬁned by

F(x

) = k exp[−(x

+ x

)/σ

] (2)

where σ is the standard deviation of the ﬁlter and con-

trols the amount of spatial detail that is retained, and

model.c

model.h

...

model.mk

system.tmf

make -f model.mk

model.exe

model

private.h

Executable C Program

Program

Model Code Custom

Makeﬁle

Generate

Code

Generate

Set Parameters

Conﬁguration

Template

Makeﬁle

Model

Simulink

User-developed model and template makeﬁle

Automated build process

make rtw.m

Figure 2: Automatic build process for Real-Time Work-

shop.

k is a normalization factor that keeps the area under

Gaussian curve equal to 1.

It is a direct result that color constancy (i.e., inde-

pendence from single source illuminant spectral dis-

tribution) is reasonably complete since the image dis-

tribution, I

(x,y), value can be expressed

(x,y) = S

(x,y)r

(x,y) (3)

where S

(x,y) is the spatial distribution of the source

illumination and r

(x,y), the distribution of scene re-

ﬂectances, so that

(x,y) = log

(x,y)r

(x,y)

(x,y)r

(x,y)

(4)

where the bars denote spatially weighted average

value. As long as S

(x,y) ≈ S

(x,y), then

(x,y) ≈ log

(x,y)

(5)

4.2 Prototyping Retinex On Ti

C6000-based Dsp

Once the retinex model was simulated by passing of

different images and its performance was checked

then the real-time C code for a speciﬁc target was

generated through Real Time Workshop. Before two

channels for RTDX communication were added. In

this process, the executable application was built and

downloaded automatically by CCS IDE in the DSP

target. At last, real-time digital data was send for dig-

ital processing through two programs based on RTDX

instrumentation. Processed images were received for

the other channel in the same way for visualization

with different tools of MATLAB.

This is not an impediment to effective use of tools

provided for Code Composer Studio IDE. Moreover,

these tools may sometimes be necessary to use, fol-

lowing the traditional practice of development of real-

time applications using Code Composer Studio IDE

exclusively. Code Composer Studio provides tools

for conﬁguring, building, debugging, tracing and an-

alyzing programs. Texas Instruments DSP’s provide

on-chip emulation support that enables Code Com-

poser Studio to control program execution and moni-

tor real-time program activity. Finally to point out that

the C code generated by this rapid prototyping plat-

form is not the most efﬁcient. To obtain a better code

and increase the performance, there are techniques

to improve and modify the C code generated. How-

ever, these techniques are not rapid and easy pricessly

because designers must manually optimize the gen-

erated code in the code Composer Studio IDE. Al-

ternatively, the code can be optimized by modifying

the corresponding blocks of Simulink model and us-

ing others blocks from the preoptimized C62x and

C64x libraries. When the code is generated, the Em-

bedded Target for TI C6000 DSP produces function

calls to preoptimized assembler implementations of

the blocks, increasing the efﬁciency and performance

of critical zones of real-time application.

5 CONCLUSION

We have successfully implemented a real-time ver-

sion of simple scale Retinex image enhancement

algorithm using a digital signal processor. This

methodology allows an easy portability a other sim-

ilar processors. We have discussed in this paper a de-

sign methodology of retinex algorithm based on cut-

ting edge like MATLAB/Simulink, Code Composer

studio IDE, and EVM/DSK hardware based on C6000

DSP of Texas Instruments. The paper described and

illustrated the most important point to consider dur-

ing the application of this technology. This technol-

ogy lets develop and validate digital signal process-

ing designs from concept through code, in a typi-

cal professional vision design simulation implemen-

tation. Moreover, the above discussion illustrated the

use of the rapid prototyping system is a fully auto-

mated program building process, where the system is

tested prior to the generation of the executable ﬁle.

This saves tremendous amount of time, besides re-

ducing the hardware cost. In the continued effort to

train more DSP engineer, this type of technology in-

corporates an added proﬁt to the formation of new ex-

perts in this knowledge area and can help to speed

up the learning curve and implementation of real-time

DSP applications. Another important beneﬁt is that it

avoids low level hardware work that can be tedious

and very time consuming and therefore designers can

focus their efforts in another important aspects of de-

sign of real-time applications. This work describes

the steps needed to write and RTDX host application

using MATLAB and the Developer’s Kit for Texas

Instruments DSP. Finally we illustrated this process

with some applications presented in this paper and its

feasibility is proved.

REFERENCES

Allensworth, D. (2002). How to Write an RTDX Host Ap-

plication using Matlab. Technical report, Texas Instru-

ments.

Assis de Melo, M.A.; Leonardi, F. and La Neve, A. (2004).

Digital Signal Processing with Matlab and DSP Kits.

In Digital Signal Processing Workshop, 2004 and

the 3rd IEEE Signal Processing Education Workshop.

2004 IEEE 11th, pages 15–18.

Chacon, M. and Valenzuela, I. (2004). Fast Image Process-

ing Application Development Scheme for the DSK

C6711 using Matlab and Simulink. In Digital Signal

Processing Workshop, 2004, pages 79–83.

Chassaing, R. (2002). DSP Applications Using C and the

TMS320C6x DSK. John Wiley & Sons, New York.

Gan, W. (2002). Teaching and Learning the Hows and Whys

of Real-time Digital Signal Processing. IEEE Trans-

actions on Education, 45(4):336–343.

Gan, W.-S. and Kuo, S. M. (2006). Teaching DSP Software

Development: from Design to Fixed-point Implemen-

tations. IEEE Transactions on Education, 49(1):122–

131.

User Guide 3 (2006). Embedded Target for TI

TMS320C6000. The MathWorks, Inc.

User Guide 6 (2006a). Link for Code Composer Studio De-

velopment Tools. The MathWorks, Inc.

User Guide 6 (2006b). Real-Time Workshop. The Math-

Works, Inc.