ImageNet classification with deep convolutional neural networks
Alex Krizhevsky, Ilya Sutskever et al.
831
Citations
44
Influential Citations
IEEE Transactions on Image Processing
Venue
1998
Year
A hybrid multidimensional image segmentation algorithm is proposed, which combines edge and region-based techniques through the morphological algorithm of watersheds. An edge-preserving statistical noise reduction approach is used as a preprocessing stage in order to compute an accurate estimate of the image gradient. Then, an initial partitioning of the image into primitive regions is produced by applying the watershed transform on the image gradient magnitude. This initial segmentation is the input to a computationally efficient hierarchical (bottom-up) region merging process that produces the final segmentation. The latter process uses the region adjacency graph (RAG) representation of the image regions. At each step, the most similar pair of regions is determined (minimum cost RAG edge), the regions are merged and the RAG is updated. Traditionally, the above is implemented by storing all RAG edges in a priority queue. We propose a significantly faster algorithm, which additionally maintains the so-called nearest neighbor graph, due to which the priority queue size and processing time are drastically reduced. The final segmentation provides, due to the RAG, one-pixel wide, closed, and accurately localized contours/surfaces. Experimental results obtained with two-dimensional/three-dimensional (2-D/3-D) magnetic resonance images are presented.
This 1998 paper addresses a fundamental challenge in image segmentation: combining the strengths of edge-based and region-based approaches while maintaining computational efficiency. The watershed transform is known for producing closed contours but suffers from over-segmentation; region merging can refine this but is often slow. By introducing a nearest neighbor graph to accelerate the merging process, the authors made hierarchical region merging practical for larger images, including 3D volumes. This work is particularly significant for medical imaging, where accurate and efficient segmentation of anatomical structures from MRI is critical.
The paper's hybrid approach remains relevant today as many modern segmentation pipelines still rely on initial over-segmentation followed by merging or classification. The idea of using graph-based acceleration to reduce priority queue operations is a classic optimization that influenced later work in superpixel segmentation and graph-based image processing.
The key innovations include:
The paper presents experimental results on 2D and 3D magnetic resonance images. The authors demonstrate that the proposed algorithm produces visually accurate segmentations with closed, well-localized boundaries. The key quantitative claim is computational efficiency: the NNG-based method drastically reduces processing time compared to the traditional priority queue approach, though exact speedup factors are not given in the abstract. No quantitative segmentation accuracy metrics (e.g., overlap with ground truth) are reported, which is a limitation.
This paper contributed to the development of efficient, hybrid segmentation methods that are still used in medical image analysis. The nearest neighbor graph acceleration technique is a classic optimization that has been adopted in various graph-based segmentation algorithms. While deep learning has largely superseded traditional methods for many tasks, the principles of over-segmentation followed by merging remain relevant in superpixel generation and interactive segmentation. The paper's focus on computational efficiency for 3D data was ahead of its time and influenced later work on scalable segmentation.
Alex Krizhevsky, Ilya Sutskever et al.
Ashish Vaswani, Noam Shazeer et al.
Douglas M. Bates, Martin Mächler et al.
Diederik P. Kingma, Jimmy Ba