The authors developed the EMDS-6: Environmental Microorganism Image Dataset Sixth Version, which contains 21 types of environmental microorganism (EMs). Each type of EM contains 40 original and 40 GT images, in total 1680 EM images. In their study, in order to test the effectiveness of EMDS-6. They choose the classic algorithms of image processing methods such as image denoising, image segmentation and target detection. The experimental result shows that EMDS-6 can be used to evaluate the performance of image denoising, image segmentation, image feature extraction, image classification, and object detection methods.
Environmental microorganisms (EMs) are minuscule life forms that exist in nature, invisible to the naked eye and detectable only through microscopic observation. Despite their size, EMs wield significant influence on human existence. Certain beneficial bacteria contribute to the production of fermented foods such as cheese and bread, offering a positive perspective. Conversely, some EMs aid in plastic degradation, industrial waste gas treatment, and soil enhancement. However, viewed in a negative light, EMs contribute to food spoilage, diminished crop yields, and serve as major contributors to the spread of infectious diseases.
To leverage the advantages of environmental microorganisms while mitigating their detrimental effects, numerous scientific researchers have dedicated themselves to studying EMs. Fundamental to this endeavor is the analysis of EM images. EMs typically range in size from 0.1 to 100 microns, presenting challenges in detection and identification. Traditional morphological methods necessitate direct microscopic examination, followed by interpretation based on shape characteristics. However, this conventional approach is labor and time-intensive. Therefore, employing computer-assisted feature extraction and analysis of EM images empowers researchers to make accurate decisions efficiently, leveraging minimal professional expertise and time investment.
Collecting samples of environmental microorganisms (EMs) typically occurs outdoors. However, during transportation to the laboratory for observation, significant environmental and temperature fluctuations can compromise the quality of EM samples. Additionally, when researchers observe EMs under a traditional optical microscope, subjective errors are common due to prolonged visual processing. Consequently, creating datasets of environmental microorganism images poses a significant challenge. Furthermore, the majority of existing EM image datasets are not publicly accessible, impeding progress in related scientific research.
EMDS-6 stands out among other environmental microorganism image datasets due to several notable advantages. Firstly, it encompasses a diverse range of microorganisms, offering ample opportunities for multi-classification of EM images. Moreover, each image within the EMDS-6 dataset is accompanied by a corresponding ground truth (GT) image. These GT images serve as valuable benchmarks for evaluating the performance of tasks like image segmentation and target detection. Furthermore, unlike many existing datasets, EMDS-6 includes GT images, enhancing its utility in various applications such as denoising, image segmentation, feature extraction, image classification, and object detection. As a result, EMDS-6 provides robust data support for a wide array of research endeavors in the field of environmental microorganisms.
An example of EMDS-6, including original images and GT images.
The EMDS-6 dataset comprises 1680 images, featuring 21 distinct classes of original EM images, each class containing 40 images. This totals 840 original images, with each accompanied by a corresponding GT image, also totaling 840. The image collection for EMDS-6 spanned from 2012 to 2020.
The authors introduced four types of noise—Poisson, multiplicative, Gaussian, and pretzel noise—into their study. By manipulating parameters such as mean, variance, and density, they generated a total of 13 specific noise types. These include multiplicative noise with variances of 0.2 and 0.04, salt and pepper noise with densities of 0.01 and 0.03 (SPN:0.01, SPN:0.03), pepper noise (PpN), salt noise (SN), Brightness Gaussian noise (BGN), Positional Gaussian noise (PGN), Gaussian noise with variances of 0.01 and mean of 0 (GN 0.01-0), Gaussian noise with variances of 0.01 and mean of 0.5 (GN 0.01-0.5), Gaussian noise with variances of 0.03 and mean of 0 (GN 0.03-0), Gaussian noise with variances of 0.03 and mean of 0.5 (GN 0.03-0.5), and Poisson noise (PN). Simultaneously, nine types of filters were applied, including the Two-Dimensional Rank Order Filter (TROF), 3×3 Wiener Filter (WF (3×3)), 5×5 Wiener Filter (WF (5×5)), 3×3 Window Mean Filter (MF (3×3)), Mean Filter with 5×5 Window (MF (5×5)), Minimum Filtering (MinF), Maximum Filtering (MaxF), Geometric Mean Filtering (GMF), and Arithmetic Mean Filtering (AMF). In the experimental setup, the 13 types of noise were added to the EMDS-6 dataset images, followed by filtering using the nine specified filters.
An example of EMDS-6, including original images and GT images.