Efficient Graph-based Image Segmentation Research Articles

Graph-Based Image Segmentation for Road Extraction from Post-Disaster Aerial Footage

This research effort proposes a novel method for identifying and extracting roads from aerial images taken after a disaster using graph-based image segmentation. The dataset that is used consists of images taken by an Unmanned Aerial Vehicle (UAV) at the University of West Florida (UWF) after hurricane Sally. Ground truth masks were created for these images, which divide the image pixels into three categories: road, non-road, and uncertain. A specific pre-processing step was implemented, which used Catmull–Rom cubic interpolation to resize the image. Moreover, the Gaussian filter used in Efficient Graph-Based Image Segmentation is replaced with a median filter, and the color space is converted from RGB to HSV. The Efficient Graph-Based Image Segmentation is further modified by (i) changing the Moore pixel neighborhood to the Von Neumann pixel neighborhood, (ii) introducing a new adaptive isoperimetric quotient threshold function, (iii) changing the distance function used to create the graph edges, and (iv) changing the sorting algorithm so that the algorithm can run more effectively. Finally, a simple function to automatically compute the k (scale) parameter is added. A new post-processing heuristic is proposed for road extraction, and the Intersection over Union evaluation metric is used to quantify the road extraction performance. The proposed method maintains high performance on all of the images in the dataset and achieves an Intersection over Union (IoU) score, which is significantly higher than the score of a similar road extraction technique using K-means clustering.

Segmentation and classification algorithms applied to sentinel-2A images for geological mapping: case of the Al Glo'a sheet (1/50000), Morocco

The multispectral image, of Landsat 7 and 8; Aster and Sentinel-2A, has good results in lithological, structural, hydrothermal and mineralogical alteration mapping. Segmentation and image classification are two complementary steps as they are allowed the most important operations in the field of image processing. In this sense, this work aims at evaluating the potential of segmentation and classification algorithms for the generation of surface geological maps, hydrothermal alterations and lineaments. Given the good resolution of the Sentinel 2A (10m), the three images, (11/12; 11/2; 11/8), 12.8.2, main component 1, 2 and 3 (11.12.2), processed by five algorithms (K-means, isodata, watershed, efficient graph-based image segmentation, thresholding) for geological mapping and then mining exploration. The study displayed that 1) Watershed algorithm can be used for topographic and hydraulic studies, it can be very useful in the preparation phase of geological and mining infrastructures; 2) threshold segmentation does not give good results in terms of geological discrimination since it divides each image into two parts; 3) the same thing for the effective thresholding and segmentation of graph-based images; 4) The Isodata and K-means algorithms show good geological discrimination.

A vision system for canning with fish sensing using rule‐based matching and segmentation

AbstractWe develop a machine vision system as a key component of a robot‐assisted packaging system, which can guide the robot arms to pack the roast sauries into cans. For gripping strategy generation, the system is required not only to be able to detect the roast saury area but also estimate the geometric parameters. Besides, according to different canning requirements, it is also necessary to distinguish the type of fish parts. Facing these challenges, we propose a novel rule‐based matching method combined with an improved efficient graph‐based image segmentation (EGIS) method for sensing the fish part. Specifically, the matching method applies our originally designed rule‐based similarity under a genetic algorithm framework combined with deterministic crowding technique, which is used to sensing one type of fish parts. On the other hand, we improve the EGIS by introducing a shape restriction to deal with leftover fish parts. The experiments are implemented for two different types of fish part in the real factory environment. The result of our method achieved a mean location accuracy of 93.5% with a practical average processing time of 2.6 s per image. © 2020 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Localization of diabetic macular edema areas via graph-based segmentation of OCT retinal images

We propose a technique for localization of diabetic macular edema areas via graph-based segmentation of OCT retinal images. The relevance of the research is associated with the high incidence rate of severe eye conditions due to diabetic macular edema among the world population. The technique relies upon a highly efficient graph-based image segmentation. Using a set of specially selected parameters, the accuracy of retinal area segmentation is enhanced. Optimal parameters found in the course of research have enabled a segmentation error of 2% to be achieved.

Local Variation as a Statistical Hypothesis Test

The goal of image oversegmentation is to divide an image into several pieces, each of which should ideally be part of an object. One of the simplest and yet most effective oversegmentation algorithms is known as local variation (LV) Felzenszwalb and Huttenlocher in Efficient graph-based image segmentation. IJCV 59(2):167–181 (2004). In this work, we study this algorithm and show that algorithms similar to LV can be devised by applying different statistical models and decisions, thus providing further theoretical justification and a well-founded explanation for the unexpected high performance of the LV approach. Some of these algorithms are based on statistics of natural images and on a hypothesis testing decision; we denote these algorithms probabilistic local variation (pLV). The best pLV algorithm, which relies on censored estimation, presents state-of-the-art results while keeping the same computational complexity of the LV algorithm.

Detection of small-sized insect pest in greenhouses based on multifractal analysis

A new application of multifractal analysis for the detection of small-sized pests (e.g., whitefly) from leaf surface images in situ is proposed in this paper. Multifractal analysis was adopted for segmentation of whitefly images based on the local singularity and global image characters with the regional minima selection strategy. According to the multifractal dimension, the candidate blobs of whiteflies were initially defined from the leaf image. The regional minima were utilized for feature extraction of candidate whitefly image areas and the performance was compared to that of the fixed threshold. Subsequently, most false alarms from leaf veins were decreased by consideration of the size and shape of the whiteflies. Experiments were conducted with field images in a greenhouse. Detection results were compared with other adaptive segmentation algorithms. Values of F measuring precision and recall scores were higher for the proposed multifractal analysis (88.6%) than for conventional methods such as Watershed (60.2%) and Efficient Graph-based Image Segmentation (EGBIS; 42.8%). The true-positive rate of multifractal analysis was 86.9% and the false-positive rate was at the minimum level of 8.2%. Overall, the detection of small-sized pests is most feasible with the proposed multifractal analysis under greenhouse conditions.

모바일 장치에서 에지 선택의 효율성

본 논문은 모바일 환경 하에서 효과적인 그래프 기반의 이미지 분할 적용 시, 실행시간 및 메모리 사용량 향상시키고, 이를 이용해 surface layout의 전처리 작업으로 수퍼 픽셀을 얻기 위한 효율적인 방법을 제안한다. 그래프기반의 이미지 분할은 영상으로부터 오브젝트 영역을 추출하는 알고리즘으로 입력 영상의 각 화소 정보를 이용한 에지 생성 및 그래프를 구성한다. 그래프 구성 시 기준 화소 및 주변 화소 위치 정보 및 색상 강도 차이 값(edgeweight)을 이용하여 에지를 구성하며, 색상 강도 차이 값에 따른 문턱치 값을 기준으로 각 에지를 연결함으로써 객체 영역을 추출하고 있다. 따라서 에지의 수는 그래프 기반의 이미지 분할의 실행시간 및 메모리 사용량을 결정하게된다. 모바일 디바이스는 PC에 비해 프로세서 속도 및 메모리 사용량 등 하드웨어적인 제약사항이 많으며, 프로그램 응답 시간이 주요 이슈가 되고 있다. 이러한 모바일 디바이스 상의 영상 처리 기술 적용 시 빠른 프로그램 응답시간 제공은 반드시 이루어져야 하며, 이미지 분할 단계에서도 보다 빠른 응답 시간을 위한 프로세싱 시간과 메모리사용량을 줄여야만 한다. 본 논문은 그래프 상의 에지의 수를 효과적으로 선택하는 효과적인 에지 선택 알고리즘 적용을 통해 그래프 기반의 이미지 분할 알고리즘을 모바일 디바이스에 적용 시, 프로세싱 시간 및 메모리 사용량의 개선을 보이도록 하고, 수퍼 픽셀 당 정확도가 입력영상 사이즈에 관계없이 70%이상 생성되며, 그 정확도가 96%동일하다는 것을 보인다. This paper proposes the effective edge selection algorithm for the rapid processing time and low memory usage of efficient graph-based image segmentation on mobile device. The graph-based image segmentation algorithm is to extract objects from a single image. The objects are consisting of graph edges, which are created by information of each image's pixel. The edge of graph is created by the difference of color intensity between the pixel and neighborhood pixels. The object regions are found by connecting the edges, based on color intensity and threshold value. Therefore, the number of edges decides on the processing time and amount of memory usage of graph-based image segmentation. Comparing to personal computer, the mobile device has many limitations such as processor speed and amount of memory. Additionally, the response time of application is an issue of mobile device programming. The image processing on mobile device should offer the reasonable response time, so that, the image segmentation processing on mobile should provide with the rapid processing time and low memory usage. In this paper, we demonstrate the performance of the effective edge selection algorithm, which effectively controls the edges of graph for the rapid processing time and low memory usage of graph-based image segmentation on mobile device.

Spatial Object Segmentation Using Stereo Images

The framework of our proposed for segmenting objects using spatial location information from stereo images. An efficient graph-based image segmentation algorithm within this framework for combining changes in optical features and physical location to segment reality scenes into perceptually and semantically uniform regions. Optical and physical location are extracted usingk-means clustering, and we propose a rules table for combining optical and spatial features together. The performance of our proposed framework is demonstrated in a series of reality-scene images using experimental data from the Middlebury stereo image data(http://vision.middlebury.edu/stereo/data/).