Template Matching
248
CHAPTER 7 Image Enhancement and Analysis
FIGURE 7.3–21 Region grown from seed location (328, 341) with threshold 0.2.
out the flower. (Performed using MATLAB Central routine regiongrowing.m provided by Dirk-Jan Kroon.)
Several seeds can be used in region growing to segment multiple objects or parts of one object. Significant interaction may be required to get the desired results. As with clustering using K-means, pixel values can be replaced by textures on small blocks to get a segmentation based on local texture in addition to pixel values. Similarly, RGB color images can be treated using a distance in the relevant color space. If segmenting for human viewing, then the more perceptually uniform color space CIELAB may be used.
7.4 OBJECT DETECTION
We now briefly look at two techniques for finding objects in images. First, we look at segmentation of a given image into a small number of regions based on a distance threshold. Then we look at a matched filtering approach to detecting objects, called template matching.
Object Segmentation
We can extend the region-growing algorithm to detect objects in color images. The basic algorithm must be modified to compute distance on the YUV or Lab coordinates. In addition, a raster scan of the pixels can detect undefined pixels in an outer loop added around the preceding algorithm. Each pass through this outer loop will define a separate region, and the threshold T will determine the number of regions
7.4 Object Detection
249
found. Additionally, some small regions are formed due to noise in the image. These must be merged into their nearest neighbors, in a YUV distance-defined region. Example 7.4–1: Object Segmentation5
In this example, an object detection strategy was applied to a frame from the Miss America color video test clip of size 360 × 288 luma or Y pixels, with the U and V pixels subsampled by 2 × 2. An appropriate Euclidean distance was formed based on the YUV pixel values. The threshold was
CHAPTER 7 Image Enhancement and Analysis
FIGURE 7.3–21 Region grown from seed location (328, 341) with threshold 0.2.
out the flower. (Performed using MATLAB Central routine regiongrowing.m provided by Dirk-Jan Kroon.)
Several seeds can be used in region growing to segment multiple objects or parts of one object. Significant interaction may be required to get the desired results. As with clustering using K-means, pixel values can be replaced by textures on small blocks to get a segmentation based on local texture in addition to pixel values. Similarly, RGB color images can be treated using a distance in the relevant color space. If segmenting for human viewing, then the more perceptually uniform color space CIELAB may be used.
7.4 OBJECT DETECTION
We now briefly look at two techniques for finding objects in images. First, we look at segmentation of a given image into a small number of regions based on a distance threshold. Then we look at a matched filtering approach to detecting objects, called template matching.
Object Segmentation
We can extend the region-growing algorithm to detect objects in color images. The basic algorithm must be modified to compute distance on the YUV or Lab coordinates. In addition, a raster scan of the pixels can detect undefined pixels in an outer loop added around the preceding algorithm. Each pass through this outer loop will define a separate region, and the threshold T will determine the number of regions
7.4 Object Detection
249
found. Additionally, some small regions are formed due to noise in the image. These must be merged into their nearest neighbors, in a YUV distance-defined region. Example 7.4–1: Object Segmentation5
In this example, an object detection strategy was applied to a frame from the Miss America color video test clip of size 360 × 288 luma or Y pixels, with the U and V pixels subsampled by 2 × 2. An appropriate Euclidean distance was formed based on the YUV pixel values. The threshold was