Image Segmentation Using Deep Learning: A Survey.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image segmentation is crucial for diverse applications like medical imaging and robotics.
• Deep learning (DL) has significantly impacted the development of advanced image segmentation techniques.
• Numerous segmentation algorithms exist, but DL models are increasingly prominent.

Purpose of the Study:

• To provide a comprehensive review of recent deep learning-based image segmentation literature.
• To analyze various DL architectures and their applications in semantic and instance segmentation.
• To discuss the strengths, challenges, and future research directions in DL for image segmentation.

Main Methods:

• Review of pioneering deep learning efforts in semantic and instance segmentation.
• Categorization of DL models including convolutional networks, encoder-decoder architectures, and generative models.
• Examination of widely used datasets and performance comparisons.

Main Results:

• Identification of key DL-based segmentation approaches such as pixel-labeling networks, encoder-decoder models, and attention mechanisms.
• Analysis of the relationships, strengths, and challenges associated with different DL segmentation models.
• Comparison of model performances across various benchmark datasets.

Conclusions:

• Deep learning models offer powerful solutions for image segmentation tasks.
• Understanding the landscape of DL approaches, datasets, and performance metrics is crucial for advancing the field.
• Future research should focus on addressing current challenges and exploring novel DL architectures for enhanced segmentation capabilities.