Under-segmentation Error Research Articles

A novel hybrid methodology integrating pixel- and object-based techniques for mapping land use and land cover from high-resolution satellite data

ABSTRACT The classification of very high-resolution satellite imagery remains a focal point in remote sensing, attracting increased attention across diverse scientific disciplines. Various classification methods, including pixel- and object-based techniques, have been proposed, and their performances and limitations have been discussed in the literature. This paper presents a hybrid method that combines the strengths of pixel- and object-based methods in image classification to minimize errors associated with the segmentation process, particularly under-segmentation errors in object-based image analysis. The core concept behind the method lies in categorizing segmented image objects as either homogeneous or heterogeneous based on their class probability. In this process, the estimated possibilities from the object-based classification model are considered, and segments are designated as homogeneous or heterogeneous using a user-defined threshold. The object-based classification model determines the class labels for homogeneous image objects, while the heterogeneous ones, containing pixels representing different land cover classes, are classified using the pixel-based model. The performance of hybrid classification models, created by varying thresholds, is analysed using high-resolution WorldView-3 and WorldView-2 imagery and compared with pixel- and object-based classification results. For the implementation of image classification methods, Canonical Correlation Forest (CCF), Extreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost) were employed. The findings indicated that employing the suggested hybrid strategy with a threshold value selected within a specific range (e.g. between 60% and 80%) and employing a robust classification algorithm that provides class probabilities (e.g. CCF) results in a statistically significant improvement in overall accuracy compared to pixel and object-based methods, with gains of 5% and 4%, respectively. Visual analysis of the produced thematic maps revealed that the proposed method minimizes the salt-and-pepper effect associated with pixel-based classification while mitigating segmentation problems arising from object-based classification.

USNet: underwater image superpixel segmentation via multi-scale water-net

Underwater images commonly suffer from a variety of quality degradations, such as color casts, low contrast, blurring details, and limited visibility. Existing superpixel segmentation algorithms face challenges in achieving superior performance when directly applied to underwater images with quality degradation. In this paper, to alleviate the limitations of superpixel segmentation when applied to underwater scenes, we propose the first underwater superpixel segmentation network (USNet), specifically designed according to the intrinsic characteristics of underwater images. Considering the quality degradation, we propose a multi-scale water-net module (MWM) aimed at enhancing the quality of underwater images before superpixel segmentation. The degradation-aware attention (DA) mechanism is then created and incorporated into MWM to solve light scattering and absorption, which can decrease object visibility and cause blurred edges. By effectively directing the network to prioritize locations that exhibit a considerable decrease in quality, this method enhances the visibility of those specific areas. Additionally, we extract the deep spatial features using the coordinate attention method. Finally, these features are fused with the shallow spatial information using the dynamic spatiality embedding module to embed comprehensive spatial features. Training and testing were conducted on the SUIM dataset, the underwater change detection dataset, and UIEB dataset. Experimental results show that our method achieves the best scores in terms of achievable segmentation accuracy, undersegmentation error, and boundary recall evaluation metrics compared to other methods. Both quantitative and qualitative evaluations demonstrate that our method can handle complicated underwater scenes and outperform existing state-of-the-art segmentation methods.

Weakly supervised learning for subcutaneous edema segmentation of abdominal CT using pseudo-labels and multi-stage nnU-Nets.

Volumetric assessment of edema due to anasarca can help monitor the progression of diseases such as kidney, liver or heart failure. The ability to measure edema non-invasively by automatic segmentation from abdominal CT scans may be of clinical importance. The current state-of-the-art method for edema segmentation using intensity priors is susceptible to false positives or under-segmentation errors. The application of modern supervised deep learning methods for 3D edema segmentation is limited due to challenges in manual annotation of edema. In the absence of accurate 3D annotations of edema, we propose a weakly supervised learning method that uses edema segmentations produced by intensity priors as pseudo-labels, along with pseudo-labels of muscle, subcutaneous and visceral adipose tissues for context, to produce more refined segmentations with demonstrably lower segmentation errors. The proposed method employs nnU-Nets in multiple stages to produce the final edema segmentation. The results demonstrate the potential of weakly supervised learning using edema and tissue pseudo-labels in improved quantification of edema for clinical applications.

The application of Hyperpixel segmentation algorithm in Chinese painting

Chinese painting, as an important component of traditional Chinese art, not only expresses Chinese culture and history, but also contains rich artistic connotations. Therefore, extracting and analyzing the main objects of Chinese painting is of great significance. The research introduces a fusion deep network based super pixel segmentation algorithm for image segmentation, and utilizes user marked segmentation target objects and backgrounds. The maximum similarity region merging idea is adopted to extract the main subject objects in Chinese painting, thereby achieving a deep understanding of the work. The results show that the boundary recall and segmentation accuracy of the Chinese painting subject object segmentation algorithm based on super pixels on the BuptPainting dataset are as high as 94.38% and 98.06%, respectively, with an undersegmentation error rate of only 7.69%. Meanwhile, the average accuracy value, optimal data scale value, and optimal image scale value of this method on the BuptPainting dataset are as high as 0.657, 0.541, and 0.529, respectively. The method proposed by the research institute has significant advantages in segmenting the main objects of Chinese painting, providing new ideas and methods for image analysis and processing of modern Chinese painting, and helping to further explore the artistic charm and cultural connotations of Chinese painting.

BCNN: Backpropagation CNN-Based fully unsupervised skull stripping for accurate brain segmentation

Brain skull stripping is an essential step before performing the segmentation. It leads to better performance and less computational load on the model due to the elimination of redundant features from the image. However, the preparation of the skull-stripped ground truth brain images by the experts is a very tedious task and may lead to human errors. In this article, a fully unsupervised approach to brain extraction has been proposed. The cascaded loss function is used and is tuned for better segmentation. The number of connected components is also tuned in each type of dataset. The cascaded loss function is a combination of focal loss (FL) and dice loss (DL). To address the class imbalance issue, the leaky ReLU activation function is used. Enhancement of the brain image has been performed before extraction which yielded better performance. In comparison with other methods, the proposed backpropagation-based convolutional neural network (BCNN) work gives better qualitative and quantitative outcomes on four out of seven parameters. The dice similarity coefficient (DSC) of the proposed model is 0.89 which is the highest as compared to the other models for brain extraction. The specificity of the model is 0.998 with 97.21 % accuracy. The undersegmentation error is reduced to 1.2002 using the cascaded loss function. The proposed work has been evaluated on four brain image datasets. It has been found that the proposed model extracts the brain from the skull and also the white matter, gray matter, and cerebrospinal fluid. Therefore, it has been noted that the proposed model will be beneficial for efficient brain skull stripping without the availability of ground truth data.

Object-based crop classification in Hetao irrigation zone by using deep learning and region merging optimization

crop classification is conducive to precision agriculture. Due to the cost of high-resolution image collection, it is uneasy to conduct crop classification in remotely sensed scenes using deep networks, which have become increasingly popular in remote sensing. This work combines geographical-based image analysis (GEOBIA) with deep learning for crop classification in a small area. An image classifier network is designed by using multi-scale CNN and transformer modules. The network input is an image transformed from a segment obtained using multi-resolution segmentation (MRS). An iterative optimization framework is developed to correct the segments with under-segmentation errors (USE). Two scenes of high-resolution images are employed for the experiment. The proposed optimization algorithm leads to superior performance to competitors. By using the proposed classifier network as a baseline, the optimization approach can improve overall accuracy (OA) by 4.33 % and 1.29 % respectively for the first and second dataset.

An Accurate and Efficient Supervoxel Re-Segmentation Approach for Large-Scale Point Clouds Using Plane Constraints

The accurate and efficient segmentation of large-scale urban point clouds is crucial for many higher-level tasks, such as boundary line extraction, point cloud registration, and deformation measurement. In this paper, we propose a novel supervoxel segmentation approach to address the problem of under-segmentation in local regions of point clouds at various resolutions. Our approach introduces distance constraints from boundary points to supervoxel planes in the merging stage to enhance boundary segmentation accuracy between non-coplanar supervoxels. Additionally, supervoxels with roughness above a threshold are re-segmented using random sample consensus (RANSAC) to address multi-planar coupling within local areas of the point clouds. We tested the proposed method on two publicly available large-scale point cloud datasets. The results show that the new method outperforms two classical methods in terms of boundary recall, under-segmentation error, and average entropy in urban scenes.

BACA: Superpixel Segmentation with Boundary Awareness and Content Adaptation

Superpixels could aggregate pixels with similar properties, thus reducing the number of image primitives for subsequent advanced computer vision tasks. Nevertheless, existing algorithms are not effective enough to tackle computing redundancy and inaccurate segmentation. To this end, an optimized superpixel generation framework termed Boundary Awareness and Content Adaptation (BACA) is presented. Firstly, an adaptive seed sampling method based on content complexity is proposed in the initialization stage. Different from the conventional uniform mesh initialization, it takes content differentiation into consideration to incipiently eliminate the redundancy of seed distribution. In addition to the efficient initialization strategy, this work also leverages contour prior information to strengthen the boundary adherence from whole to part. During the similarity calculation of inspecting the unlabeled pixels in the non-iterative clustering framework, a multi-feature associated measurement is put forward to ameliorate the misclassification of boundary pixels. Experimental results indicate that the two optimizations could generate a synergistic effect. The integrated BACA achieves an outstanding under-segmentation error (3.34%) on the BSD dataset over the state-of-the-art performances with a minimum number of superpixels (345). Furthermore, it is not limited to image segmentation and can be facilitated by remote sensing imaging analysis.

Automated extraction of aquaculture ponds from Sentinel-2 seasonal imagery – A validated case study in central Thailand

The incidence and size of aquaculture ponds are related to a variety of economic, social, environmental, and policy factors, but there is scarce publicly available information. Satellite based aquaculture pond mapping has typically been undertaken by segmentation of single-date images or of temporal composites derived from image time series. In Asia, both aquaculture and rice farming can be undertaken in the same locality and, combined with frequent cloud cover, meaning that segmentations should be derived throughout the year to be able to differentiate aquaculture ponds from rice paddies and to provide spatially-complete pond mapping results. In this paper, a new approach is presented to extract individual aquaculture ponds from seasonal Sentinel-2 10 m images and to combine different sets of pond objects extracted from different images into a single set of pond objects. The approach is demonstrated for the province of Nakhon Pathom (216,800 ha) in central Thailand that is a major shrimp farming area. Aquaculture ponds were extracted from near-cloud-free Sentinel-2 images acquired in January and June 2019 to reduce confusion with rice paddies that are typically vegetation covered in these months. The ponds extracted at different times were then combined using a multi-temporal object combination strategy. Surveys undertaken in 2019 to elucidate farmers’ attitudes and land use practices were used to contextualize the extraction results. Across the province, 22,833 aquaculture ponds were extracted with a total area of 18,066 ha. The mean and median pond sizes were 0.79 ha and 0.60 ha, respectively, which is close to the 0.65 ha mean shrimp pond size in Thailand reported by other researchers via independent surveys. Current methods to map aquaculture ponds have not typically reported object-based validation results and so it is unknown whether object-level information (e.g., number of ponds and pond sizes) are reliable. Therefore, the extraction results were evaluated quantitatively using object-based accuracy metrics. A total of 1733 aquaculture ponds at three validation sites were manually digitized using multi-date GoogleEarth high-resolution images, and compared with the extracted ponds. The evaluation results indicated a robust pond extraction performance, with 87.7% object-based overall accuracy considering all the reference data and the extracted ponds at the three sites. 0.9% of the extracted ponds were over-extracted (commission error), 0.9% were under-split, 2.6% were over-split, and 6.9% of the of the reference ponds were not extracted (omission error). In addition, the mean over-segmentation error of 0.154, mean under-segmentation error of 0.111, and mean F-score (dice coefficient) of 0.858 were obtained. The causes of these errors were examined and discussed with recommended potential research using other sensor data.

POSE-FCN SUPERPIXEL SEGMENTATION FOR BUILDING FACADES BASED ON 2D TEXTURE AND 3D LOCAL POSE-VARIED SEMANTIC FEATURES

Abstract. The extraction of building facades based on image sequences has a great contribution to the construction of digital realistic cities. The superpixel segmentation algorithm is a pre-processing tool for segmentation because of its advantages of fast speed, universality and great accuracy. However, the 2D features are less reliable because building facades usually have complex texture and geometric feature. It is difficult to obtain accurate detail information of the façades by clustering the superpixels. Moreover, the process of acquiring building image sequences is easily disturbed by environmental factors, which also leads to the poor results of the superpixel segmentation. In this paper, 3D local pose-varied semantic features of buildings are defined for this problem, which are computed by 3D point clouds generated from multi-view images of buildings based on SfM and PMVS. Then, multi-modal superpixels with integration of 2D texture and 3D pose-varied semantic features are computed by using fully convolutional networks. The new method is compared with traditional superpixel segmentation method by standard superpixel segmentation result evaluation metrics such as achievable segmentation accuracy , boundary recall, and undersegmentation error. The method achieve accurate segmentation results and effectively exclude the influence of complex texture and environmental factors. In summary, The multi-modal superpixels obtained by the integration have better reliability and provide a new idea for the superpixel segmentation of building facades, which has important theoretical and practical significance.

Correcting the Results of CHM-Based Individual Tree Detection Algorithms to Improve Their Accuracy and Reliability

Individual tree detection algorithms (ITD) are used to obtain accurate information about trees. Following the process of individual tree detection, it is possible to use additional processing tools to determine tree parameters such as tree height, crown base height, crown volume, or stem volume. However, many of the methods developed so far have focused on parameterising the algorithms based on the study area, height structure or tree species analysed. Applying the parameters of the method can be challenging in areas with dense and heterogeneous forests with a diverse stand structure. Therefore, this work aimed to develop a method to correct the results of ITD algorithms to identify individual trees more reliably, taking into account different ITD methods based on the Canopy Height Model. In the present study, we proposed a three-step approach to correct segmentation errors. In the first step, erroneous (under- and over-segmentation errors) and correct segments were classified. After classification, the second step was to refine the under-segmentation errors. The final step was to merge segments from the over-segmentation class with correct segments based on the specified conditions. The study was conducted in one of the most complex and diverse forest communities in Europe, making tree identification a major challenge. The accuracy of the segmentation improvements varied depending on the method applied and tree species group examined. Thus, based on the results, the paper advocates for the correction method due to its efficiency in mixed forest stands. Therefore, the present study offers a possible solution to reduce segmentation errors by considering different forest types and different CHM-based ITD methods for identifying individual trees.

A segmentation algorithm incorporating superpixel block and holistically nested edge for sugarcane aphids images under natural light conditions

Image-based automatic detection systems illustrate the efficient capability of the pest ( e.g. , sugarcane aphids) identification. However, most of the algorithms were developed using the images captured under stable light conditions and the natural illumination has been rarely considered. Here, we developed a fusion superpixel method and compared it with eight other state-of-the-art superpixel algorithms on a developed dataset of images under various natural light conditions of strong light, diffuse light, weak light, and direct sunlight. We evaluated the algorithm performance according to visual quality, ground truth deviation ( e.g. , target compactness, regularity index, boundary recall, and under-segmentation error), and segmentation runtime and quality. We found 400 as the best number of superpixels for segmenting the pest image with the resolution of 408 × 306 pixels and captured with the distance of ∼0.2 m between target and camera lens. Compared with others, the developed fusion algorithm illustrated a relatively better performance in target compactness (0.61), boundary recall (0.76), regularity index (16.4), and runtime (0.13 s). Furthermore, diffuse light was the ideal light condition in the sugarcane aphids’ identification task. Our study suggested the developed algorithm as an appropriate method concerning various natural light conditions and pest distribution in the field. • A Holistically-Nested Edge was introduced in superpixel for image segmentation. • The proposed superpixel algorithm performed better compared with previous methods. • Diffuse light was the ideal light condition in pest identification task. • K = 400 is the best number of superpixels for segmenting our sugarcane aphid images.

A machine learning approach for identifying and delineating agricultural fields and their multi-temporal dynamics using three decades of Landsat data

High spatial and temporal resolution satellite imagery are essential data for land cover discrimination and mapping of vegetation dynamics, offering insights into the number, extent, and condition of agricultural fields. However, an accurate account of the number, location, and variability of fields can be challenging to obtain in a timely manner, particularly at scale or in regions where ground-supporting data are not available, limiting the capacity for food production and water use assessment and planning. To bridge this capacity gap, a convolutional neural network approach was adopted and combined with two clustering techniques: the density-based spatial clustering of applications with noise (DBSCAN) and spectral clustering, to provide an account of agricultural fields and their extent across a region lacking ground based information. A random forest classification was also employed to discriminate crop types. Using an annual maximum normalized difference vegetation index (NDVI) derived from 2018 Landsat-8 data, the approach was applied to classify the shape and delineate the extent of agricultural fields across an agricultural region in Saudi Arabia that had an area under irrigation exceeding 2,300 km2. When assessed against manually identified center-pivot fields (CPFs), the method achieved 97.4% producer’s and 98.0% user’s accuracies on an object basis, and 81.4% producer’s and 85.4% user’s accuracies on a pixel basis for identifying non-CPFs (i.e., tree crop plantations and other non-woody crops). The over- and under-segmentation error for CPFs was 1.5% and 1.0%, respectively, with intersection over union errors reported as being 3.5%. The framework showed stability when retrospectively applied to Landsat data from the year 2000, returning 97.5% producer’s and 96.6% user’s accuracies for CPF identification. In order to characterize the temporal dynamics of agricultural development over the past three decades, an analysis of field behavior between 1988 and 2020 was subsequently undertaken. The analysis indicated that the number of CPFs in the study region increased from 45 (covering 20 km2) in 1988 to 5,080 CPFs by 2016 (covering 2,368 km2), followed by a recent reduction to 3,700 CPFs in 2020 (covering 1,581 km2). Through the multi-temporal analysis, individual fields were able to be characterized in terms of their expansion, contraction and activity throughout the study period. Overall, the proposed method was simple to train, efficient in dealing with large datasets, relied on limited in-situ records to a very small degree, and has the potential to be applied to larger national scales, providing an ongoing assessment of important agroinformatic metrics.

Local Scale-Guided Hierarchical Region Merging and Further Over- and Under-Segmentation Processing for Hybrid Remote Sensing Image Segmentation

With the development of medium- and high-resolution satellites, successfully segmenting differently sized geo-objects remains a challenging issue in the framework of geographic object-based image analysis (GEOBIA). The hybrid image segmentation method is a good alternative to produce good segmentation that best matched the different sizes of geo-objects. However, the existing methods almost use a certain scale or other segmentation parameters (SPs) to control the sizes and shapes of segments. This will lead to two issues: (1) one single scale is impossible to segment every geo-object well due to the land cover complexity within remote-sensing imageries; (2) over- and under-segmented regions still occur in the segmentation results, whatever using any advanced segmentation methods. To solve the above problems, this paper developed a hybrid image segmentation method with local scale-guided hierarchical region merging and further over- and under-segmentation processing. First, the primitive segmentation was produced and then stratified into layers with different land covers. Then, the local scale was calculated for a more objective merging process in the separating layers. Third, the over- and under-segmentation at separating layers was recognized and re-processed for achieving a fine segmentation. To validate the proposed method, it was applied to three test images of gaofen-1 satellite with different land cover types, and ten competing methods were compared. The visual and quantitative results indicated the advantage of our method in segmenting out different sizes of geo-objects, which can effectively reduce the over- and under-segmentation error.

A Supervoxel Segmentation Method With Adaptive Centroid Initialization for Point Clouds

Supervoxels find applications as a pre-processing step in many image processing problems due to their ability to present a regional representation of points by correlating them into a set of clusters. Besides reducing the overall computational time for subsequent algorithms, the desirable properties in supervoxels are adherence to object boundaries and compactness. Existing supervoxel segmentation methods define the size of a supervoxel based on a user inputted resolution value. A fixed resolution results in poor performance in point clouds with non-uniform density. Whereas, other methods, in their quest for better boundary adherence, produce supervoxels with irregular shapes and elongated boundaries. In this article, we propose a new supervoxel segmentation method, based on k-means algorithm, with dynamic cluster seed initialization to ensure uniform distribution of cluster seeds in point clouds with variable densities. We also propose a new cluster seed initialization strategy, based on histogram binning of surface normals, for better boundary adherence. Our algorithm is parameter-free and gives equal importance to the color, spatial location and orientation of the points resulting in compact supervoxels with tight boundaries. We test the efficacy of our algorithm on a publicly available point cloud dataset consisting of 1449 pairs of indoor RGB-D images, i.e., color (RGB) images coupled with depth information (D) mapped per pixel. Results are compared against three state-of-the-art algorithms based on four quality metrics. Results show that our method provides significant improvement over other methods in the undersegmentation error and compactness metrics and, performs equally well in the boundary recall and contour density metrics.

Multi-density peaks clustering superpixel

A superpixel segmentation algorithm called multi-density peaks clustering (MDPC) is proposed. By selecting a sufficient number of local density maximum pixels from the image as cluster centers to depict the image texture, the boundary of the object can be captured very accurately. The algorithm framework of MDPC is divided into three steps. First, the local density of pixel is defined, and the local density maximum pixels are calculated. Then, the local density maximum pixels are used as cluster centers, and the global optimal search, which is based on the path-to-point idea, is used to complete the clustering of the remaining non-cluster center pixels to realize the initial segmentation. Finally, superpixels are obtained by merging the initial segments according to the size of the segments and the distance between adjacent segments. In quantitative comparisons, MDPC is compared with 13 state-of-the-art superpixel segmentation algorithms in three image segmentation datasets. The experimental results show that MDPC achieves better performance in terms of boundary recall, boundary precision, achievable segmentation accuracy, undersegmentation error, and explained variation. And the qualitative comparisons show that the proposed algorithm has obvious advantages over other superpixel segmentation algorithms in image detail description and boundary adherence. Finally, the practicability and stability of MDPC are further demonstrated by the application of image segmentation. The source code of MDPC will be available at https://github.com/zhaojianaaa.

A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction.

Automatic cell segmentation and tracking enables to gain quantitative insights into the processes driving cell migration. To investigate new data with minimal manual effort, cell tracking algorithms should be easy to apply and reduce manual curation time by providing automatic correction of segmentation errors. Current cell tracking algorithms, however, are either easy to apply to new data sets but lack automatic segmentation error correction, or have a vast set of parameters that needs either manual tuning or annotated data for parameter tuning. In this work, we propose a tracking algorithm with only few manually tunable parameters and automatic segmentation error correction. Moreover, no training data is needed. We compare the performance of our approach to three well-performing tracking algorithms from the Cell Tracking Challenge on data sets with simulated, degraded segmentation-including false negatives, over- and under-segmentation errors. Our tracking algorithm can correct false negatives, over- and under-segmentation errors as well as a mixture of the aforementioned segmentation errors. On data sets with under-segmentation errors or a mixture of segmentation errors our approach performs best. Moreover, without requiring additional manual tuning, our approach ranks several times in the top 3 on the 6th edition of the Cell Tracking Challenge.

Automatic Segmentation of Digital Pathology Slides Based on Unsupervised Learning

To segment images through an unsupervised method as an alternative to manual labeling. A total of 100 whole slide image (WSI) data of HE stained and Pap stained slides were selected as the research and test objects, including 70 breast slides, 20 lung slides and 10 thyroid slides. In order to ensure the diversity of data, the breast slides included those of normal tissue, inflammation and tumor, the lung slides were mainly neoplasms in the lower lobe, including those of inflammation and tumor, and the thyroid slides were of cells, all benign, obtained through fine needle aspiration. The maximum total magnification (original magnification) of each image was 400 times, and the file format was NDPI. Each WSI was manually labeled, and the labeled area of each WSI was more than 10 fields of vision. The labeled information was to be used for validity verification. An unsupervised image segmentation technique based on superpixel and fully convolution neural network algorithms was constructed and used to segment any region of interest (ROI) of unlabeled WSI. In comparison with the region adjacency graph merging method, the segmentation effect of the two methods was assessed with the under segmentation error, the boundary recall and the mean Intersection-over-Union, and the efficiency of the two methods was also compared. In the comparison of execution efficiency, the test process included the preprocessing time of superpixel, and excluded the time of loading the deep learning engine. Unsupervised automatic segmentation was implemented for any ROI region of WSI according to the texture and color. The results of the breast slides, lung slides and thyroid slides showed slight differences, and multiple tests yielded stable results. However, the performance of this method in differentiating inflammation and tumor was average. The under-segmentation error, the boundary recall and the mean Intersection-over-Union were 19.10%, 82.06% and 45.06%, respectively. The under segmentation error, the boundary recall and the mean Intersection-over-Union for the region adjacency graph merging method were 21.52%, 78.39% and 44.81%, respectively. The average time consumption of the whole process was 0.27 s in GPU mode and 1.30 s in CPU mode. The average time consumption of the region adjacency graph merging method was 10.5 s in CPU mode because the method of region adjacency graph merging was not realized in the GPU mode. This method produced ideal pixel level labeling results through simple human-computer interaction, which could effectively reduce the cost of digital pathology slide data labeling. Compared with the region adjacency graph merging method, this method had better performance in processing image texture and had faster processing speed.

Graph-Theoretic Post-Processing of Segmentation With Application to Dense Biofilms.

Recent deep learning methods have provided successful initial segmentation results for generalized cell segmentation in microscopy. However, for dense arrangements of small cells with limited ground truth for training, the deep learning methods produce both over-segmentation and under-segmentation errors. Post-processing attempts to balance the trade-off between the global goal of cell counting for instance segmentation, and local fidelity to the morphology of identified cells. The need for post-processing is especially evident for segmenting 3D bacterial cells in densely-packed communities called biofilms. A graph-based recursive clustering approach, m-LCuts, is proposed to automatically detect collinearly structured clusters and applied to post-process unsolved cells in 3D bacterial biofilm segmentation. Construction of outlier-removed graphs to extract the collinearity feature in the data adds additional novelty to m-LCuts. The superiority of m-LCuts is observed by the evaluation in cell counting with over 90% of cells correctly identified, while a lower bound of 0.8 in terms of average single-cell segmentation accuracy is maintained. This proposed method does not need manual specification of the number of cells to be segmented. Furthermore, the broad adaptation for working on various applications, with the presence of data collinearity, also makes m-LCuts stand out from the other approaches.

Superpixels With Content-Adaptive Criteria.

Superpixels are widely used in computer vision applications. Most of the existing superpixel methods use established criteria to indiscriminately process all pixels, resulting in superpixel boundary adherence and regularity being unnecessarily inter-inhibitive. This study builds upon a previous work by proposing a new segmentation strategy that classifies image content into meaningful areas containing object boundaries and meaningless parts that include color-homogeneous and texture-rich regions. Based on this classification, we design two distinct criteria to process the pixels in different environments to achieve highly accurate superpixels in content-meaningful areas and keep the regularity of the superpixels in content-meaningless regions. Additionally, we add a group of weights when adopting the color feature, successfully reducing the undersegmentation error. The superior accuracy and the moderate compactness achieved by the proposed method in comparative experiments with several state-of-the-art methods indicate that the content-adaptive criteria efficiently reduce the compromise between boundary adherence and compactness.