라벨링 개체 병합 방식을 이용한 컨테이너 코드 영역 추출 비교 연구 (A Comparative Study on Extracting Container Code Region by Merging Labeling Objects)

서지정보

· 발행기관 : 한국해운물류학회
· 수록지 정보 : 해운물류연구 / 32권 / 1호 / 49 ~ 67페이지
· 저자명 : 최창훈, 백승관

초록

본 논문에서는 컨테이너 영상 인식 시스템의 처리 과정에 컨테이너 코드 영역 추출과정이 있는데 본 논문에서는 이를 모폴로지(Morphology) 연산과 라벨링 개체 병합 방식을 이용하여 추출하는 방법을 제시하고자 한다. 입력된 영상에서 우선 영상 전체 비율을 기반으로 하여 영역 분할을 적용하여 코드부분이 있을만한 부분만을 남기고 영상에서 컬러정보를 제거하여 그레이 영상으로 변환한다. 그리고 4가지의 모폴로지(Morphology) 연산을 적용하여 서로 다른 4개의 영상을 생성한다. 생성된 4개의 영상을 라벨링하여 개체를 추출해내고 영상과 개체간의 비율을 고려하여 잡음을 제거한다. 이후 라벨링된 개체들을 병합하여 컨테이너 코드 영역을 추출해낸다. 실험 결과 유효영역 비율이 단순 영역 추출 방식을 적용한 경우 보다 높게 나타났다.

영어초록

Container code recognition is a technology that uses cameras to automatically recognize and separate the container codes engraved on the front, side, rear, and top surfaces of containers entering the container treatment plant. Often, the code is not visible due to movement or contact between containers and even accurate RFID systems fail in recognizing the code. Matching the RFID inputs and the actual codes requires time and labor; however, automating this operation using camera recognition can save both time and money, as extracting the container code region is an important intermediate process.
This study proposes an extraction technique for container code regions, which can be used in the container image recognition system. The container code is painted in different formats for each container; therefore, pre-processing when extracting the container code region is extremely important. The first pre-processing step is applying the container image to the region division. We calculate the ratio of the total image size and the vertical rod size of the container from the input container image. Further, we apply the region division by using the calculated ratio and remove the vertical rods in the input container image to leave only the image part of the code. Then, the container image region division is converted into a gray image to apply the morphology operation.
Binarization—through the process of analyzing the image—may be less efficient in the following code recognition process. Therefore, applying the four operations of morphology highlights the code image of the container; however, the image is not uniform as it is obtained outdoors. We do not use a global thresholding method, as it does not generate a clear container binary image. Instead, we use an adaptive thresholding method to determine the binary value using the threshold values of each part of the container image. Further, the labeling technique is applied to each of the four binary images to extract the object and we remove noise in the images by considering the ratio between the image and the labeling objects. We store the labeling objects in a linked list and check whether each labeling object overlaps by examining the coordinates of objects stored in the linked list. Finally, we generate the container code region through the code recognition input by merging the overlapping objects.
The number of valid regions through our technique is higher than the simple extraction method and might be effective in next generation of container code recognition systems.

참고자료

· 없음