圖片分割外文翻譯--一種可重構(gòu)的嵌入式彩色圖像分割處理軟件系統(tǒng)（英文）

1、An embedded software-reconfigurable color segmentation architecture for image processing systemsGrigorios Chrysos a,?, Apostolos Dollas a, Nikolaos Bourbakis a,ba Technical University of Crete, ECE Dept., Chania, Crete,

2、Greece b Wright State University, Engr. College ATR Center, Dayton, OH 45435, USAa r t i c l e i n f oArticle history:Available online 17 December 2011Keywords:Reconfigurable architecturesImage segmentationEmbedded syste

3、msa b s t r a c tImage segmentation is one of the first important and difficult steps of image analysis and computer visionand it is considered as one of the oldest problems in machine vision. Lately, several segmentatio

4、nalgorithms have been developed with features related to thresholding, edge location and region growingto offer an opportunity for the development of faster image/video analysis and recognition systems. Inaddition, fuzzy

5、-based segmentation algorithms have essentially contributed to synthesis of regions for bet-ter representation of objects. These algorithms have minor differences in their performance and they allperform well. Thus, the

6、selection of one algorithm vs. another will be based on subjective criteria, or, drivenby the application itself. Here, a low-cost embedded reconfigurable architecture for the Fuzzy-like reason-ing segmentation (FRS) met

7、hod is presented. The FRS method has three stages (smoothing, edge detectionand the actual segmentation). The initial smoothing operation is intended to remove noise. The smootherand edge detector algorithms are also inc

8、luded in this processing step. The segmentation algorithm usesedge information and the smoothed image to find segments present within the image. In this work theFRS segmentation algorithm was selected due to its proven g

9、ood performance on a variety of applications(face detection, motion detection, Automatic Target Recognition (ATR)) and has been developed in a low-cost, reconfigurable computing platform, aiming at low cost applications.

10、 In particular, this paper presentsthe implementation of the smoothing, edge detection and color segmentation algorithms using StretchS5000 processors and compares them with a software implementation using the Matlab. Th

11、e new architec-ture is presented in detail in this work, together with results from standard benchmarks and comparisonsto alternative technologies. This is the first such implementation that we know of, having at the sam

12、e timehigh throughput, excellent performance (at least in standard benchmarks) and low cost.? 2011 Elsevier B.V. All rights reserved.1. Introduction1.1. SegmentationMany computer vision, pattern recognition, image analys

13、is and object extraction systems have been developed during the last thirty years. At the same time, fuzzy and semi-fuzzy clustering algorithms have been also presented for the extraction and recog- nition of an object’s

14、 features. In order for these systems and algo- rithms to be successful they generally have to start with a robust smoothing and/or segmentation technique. Thus, image segmenta- tion is an important starting step for alm

15、ost all vision and pattern recognition methodologies. Several studies have been done to cat- egorize segmentation into classes based on characteristics, such as thresholding or clustering, edge detection, region growing/

16、mergingand others [1–3]. In particular, Lee and Chung [4] showed that thresholding would usually produce good results in bimodal images only, where the images comprise of only one object and its background. However, when

17、 the object area is small compared to the background area, or when both the object and background have a broad range of gray levels, selecting a good threshold is dif- ficult. Another weakness of this technique occurs wh

18、en multiple objects are present within the image. In such cases, finding sharp valleys within the histogram is further complicated, and segmenta- tion results may be very poor. Edge detection is another approach associat

19、ed to image segmentation [5]. An edge is defined as a loca- tion where a sharp change in gray level or color is detected. How- ever, in this method it is difficult to maintain the continuity of the detected edges; a segm

20、ent must always be enclosed by a continu- ous edge. Region growing or merging is a third approach for image segmentation [6]. In this case, large, easy to find continuous re- gions or segments are detected first. Afterwa

21、rds, small regions may be merged by using homogeneity criteria [7,8]. One disadvan- tage of region growing and merging is the inherently sequential0141-9331/$ - see front matter ? 2011 Elsevier B.V. All rights reserved.d

22、oi:10.1016/j.micpro.2011.12.004? Corresponding author.E-mail addresses: chrysos@mhl.tuc.gr (G. Chrysos), dollas@mhl.tuc.gr (A. Dollas),nikolaos.bourbakis@wright.edu (N. Bourbakis).Microprocessors and Microsystems 36 (201

23、2) 215–231Contents lists available at SciVerse ScienceDirectMicroprocessors and Microsystemsjournal homepage: www.elsevier.com/locate/micpro2. The FRS segmentation methodologySegmentation is a process used to facilitate

24、the extraction of objects that form an image. The FRS methodology, which is studied in this paper, consists of three steps (prior to the recognition it- self): smoothing, edge detection and color segmentation. The data-

25、flow of segmentation process is described in Fig. 1. In this work, as will be shown below, the HIS (hue, intensity, saturation) model is used, from original RGB images, an approach which is quite typical and has been sho

26、wn in literature (see Section 1) to work well.2.1. Smoothing algorithmThe images contain noise introduced either by the camera or because of the image’s transmission over a noisy medium. In either case, the noise must be

27、 removed before any further image process- ing is applied. The most common way of noise removal is the use of filters. An important concept for a smoothing algorithm is the neighborhood between two pixels. This algorithm

28、 allows for a fuz- zy degree of neighborhood, in which for each neighboring pixel there is the corresponding degree of neighborhood, as shown in Fig. 2. Each pixel’s color is compared with the color of each of its neighb

29、oring blocks, as shown in Fig. 3. The size of blocks for our implementation was 3 ? 3, which results to a strong smoothing of the image. The average color for each of the neighboring blocks was calculated taking into acc

30、ount the neighborhood membership function as shown in the Eq. (1). For smoothing, the color contrast between the center pixel and all of the surrounding blocks must be measured. The color contrast between the pixel (i, j

31、) and the block b is the Euclidean distance in the RGB domain as shown in the fol- lowing equation:Ci;j;b ¼Ppþ3 q¼p Pkþ3 s¼klsq ? Csq Ppþ3 q¼p Pkþ3 s¼klsq ð1ÞContras

32、ti;j;b ¼ ðR2 ? R1Þ2 þ ðG2 ? G1Þ2 þ ðB2 ? B1Þ2 ð2ÞThe steps of the smoothing algorithm that were implemented in this work are shown in Fig. 4 and they are presented a

33、nalyticallyin [23].2.2. Edge detectionEdge detection is the process of the limit specification of the ob- jects an image consists of. Hue, Intensity and Saturation (repre- sented as h, i, and s respectively) are one set

34、of parameters that are used to evaluate pixels’ edge strength within images. These parameters are computed from the original image RGB values by the equations below:x ¼ 0:49r þ 0:31g þ 0:2b; y ¼ 0:177

35、r þ 0:812g þ 0:011b;z ¼ 0:01g þ 0:99bl ¼ 116 yY0? ?1 3 ; a ¼ 500 xX0? ?1 3 ? yY0? ?1 3 “ #;b ¼ 200 yY0? ?1 3 ? zZ0? ?1 3 “ #i ¼ l; s ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiff

36、iffi ffia2 þ b 2 q; h ¼ tan?1 ba? ?SMOOTHING EDGE DETECTIONCOLOR SEGMENTATIONOriginal Image Smoothed ImageEdge Image Color Segmented ImageFig. 1. The dataflow of FSR algorithm.Fig. 2. Table of neighborhood