Sungkee Lee

Adaptive image segmentation using a genetic algorithm

Image segmentation is an old and difficult problem. One of the fundamental weaknesses of current computer vision systems to be used in practical applications is their inability to adapt the segmentation process as real-world changes occur in the image. We present the first closed loop image segmentation system which incorporates a genetic algorithm to adapt the segmentation process to changes in image characteristics caused by variable environmental conditions such as time of day, time of year, clouds, etc. The segmentation problem is formulated as an optimization problem and the genetic algorithm efficiently searches the hyperspace of segmentation parameter combinations to determine the parameter set which maximizes the segmentation quality criteria. The goals of our adaptive image segmentation system are to provide continuous adaptation to normal environmental variations, to exhibit learning capabilities, and to provide robust performance when interacting with a dynamic environment. We present experimental results which demonstrate learning and the ability to adapt the segmentation performance in outdoor color imagery.

Adaptive image segmentation using genetic and hybrid search methods

This paper describes an adaptive approach for the important image processing problem of image segmentation that relies on learning from experience to adapt and improve the segmentation performance. The adaptive image segmentation system incorporates a feedback loop consisting of a machine learning subsystem, an image segmentation algorithm, and an evaluation component which determines segmentation quality. The machine learning component is based on genetic adaptation and uses (separately) a pure genetic algorithm (GA) and a hybrid of GA and hill climbing (HC). When the learning subsystem is based on pure genetics, the corresponding evaluation component is based on a vector of evaluation criteria. For the hybrid case, the system employs a scalar evaluation measure which is a weighted combination of the different criteria. Experimental results for pure genetic and hybrid search methods are presented using a representative database of outdoor TV imagery. The multiobjective optimization demonstrates the ability of the adaptive image segmentation system to provide high quality segmentation results in a minimal number of generations.<<ETX>>

Closed-loop adaptive image segmentation

A closed-loop image segmentation system that incorporates a genetic algorithm to adapt the segmentation process to changes in image characteristics caused by variable environmental conditions is presented. The genetic algorithm efficiently searches the hyperspace of segmentation parameter combinations to determine the parameter set which maximizes the segmentation quality criteria. A summary of the experimental results that demonstrates the ability to perform adaptive image segmentation and to learn from experience using a collection of outdoor color imagery is given.<<ETX>>

Refocused Recognition Of Aerial Photographs At Multiple Resolution

While performing the photo interpretation task using very high resolution images, the resolution of the image is often reduced to make its processing feasible. However, in low resolution images, it becomes quite difficult to segment and locate targets of interest such as aircraft, which are relatively small. Further, in recognizing aircraft, it is generally assumed that aircraft are already located. The emphasis is placed on model matching for recognizing isolated aircraft. However, locating potential areas in the images, where aircraft may be found, is non-trivial since it requires an accurate labeling of an image. We have developed a Knowledge-Based Photo Interpretation (KEPI) system that analyzes high resolution images. This system locates aircraft by first finding large structures in low resolution images and focusing attention on areas such as tarmacs, runways, parking areas, that have high probability of containing aircraft. Higher resolution images of the regions that are the focus of attention are used in subsequent analysis. The system makes extensive use of contextual knowledge such as spatial and locational information about airport scenes. We show results using high resolution TV data.

Baseline Adaptive Image Segmentation Using a Genetic Algorithm

Genetic algorithms can be used in three different ways to provide an adaptive behavior within a computer vision system [23]. The simplest approach is to allow the genetic algorithm to modify a set of control parameters that affect the output of an existing computer vision program. By monitoring the quality of the resulting program output, the genetic algorithm can dynamically change the parameters to achieve the best performance. A second approach allows the genetic algorithm to modify the complex data structures within an algorithm or production rule system for a computer vision application. By modifying the control mechanism or agenda in an algorithm or the organization of data frames in a rule-based system, the genetic algorithm can bring about changes in the system’s behavior. Finally, the most complex implementation of an adaptive computer vision system allows the genetic algorithm to actually make changes in the executable code of a program. In most of these cases, the adaptation involves changing the condition/action statements of the rules in a production system. Since almost every image segmentation algorithm has parameters to control the segmentation results, we have adopted the first strategy listed above for this research.

Image segmentation Techniques

A wide variety of techniques are used for image segmentation [29, 35, 36, 55]. They include edge detection [20], region splitting [49, 50, 53], region merging [36, 55], clustering [19, 46], surface fitting [55], rule-based expert systems [47], relaxation [4, 6, 8, 9, 55], and integrated techniques [27, 35, 39, 55]. In this chapter, we first briefly discuss techniques based on edge detection, and region splitting and region growing, and then present the details of the Phoenix image segmentation algorithm [41, 61] that has been used in this research.

Simultaneous Optimization of Global and Local Evaluation Measures

The adaptive image segmentation systems, which are presented in Chapters 4 and 8, are designed for the optimization of a single objective function, which is a combined segmentation quality measure derived from equally weighted sum of the five quality measures. There are times, however, when it is not practical or not acceptable to simply combine global segmentation quality measures (which evaluate segmentation quality of the whole image) with local segmentation quality measures (which evaluate segmentation quality of the regions of interest). Depending upon the object detection and segmentation techniques which are used [5] and the domain of application, such as automatic target recognition, photointerpretation, autonomous vehicle navigation, etc., we may be concerned with the segmentation of only the regions of interest or the whole image.

Hybrid Search Scheme for Adaptive Image Segmentation

In this chapter, we present the adaptive image segmentation system in which a hybrid search scheme replaces the learning component used in the baseline system described in Chapter 4. The hybrid search technique combines genetic algorithms with a hill climbing technique. Experimental results are provided that compare the performance achieved in these two systems, the baseline adaptive image segmentation system and the adaptive image segmentation system with a hybrid search scheme.

Basic Experimental Results – Outdoor Imagery

In this chapter, we focus on the indoor imagery experiments and compare the performance of the baseline adaptive image segmentation system with other techniques commonly used in the Computer Vision field. These experiments are designed to test the adaptive capabilities of the segmentation system on a controlled set of images in which we constrain the elements of the scene as well as the environmental conditions. The variation between images is limited to changes in the lighting intensity for these experiments, the position of the light source remains constant. In the next chapter, we will present experiments on outdoor color imagery.

Segmentation as an Optimization Problem

In the previous chapters, we highlighted some of the characteristics of the image segmentation problem, such as the size of the parameter space, the complexity of the objective function, and variations in the objective function caused by changes in the imagery as well as the accepted definition of the function itself. In this chapter, we formulate segmentation as a parameter optimization problem and discuss the choice of genetic algorithms for adaptive image segmentation.