2012 - TPAMI - Slic superpixels compared to state-of-the-art superpixel methods

SLIC SUPERPIXELS

Algorithm

SLIC，唯一的一个参数就是 k，想要的 superpixel 的数量
To produce roughly equally sized superpixels, the grid interval is $S= \sqrt{N/k}$
- 最好当然是每个 superpixel 像素数都是一样的咯，是不是能够这么保证呢？拭目以待
怎么确定 center？
- 这其实是算法的第一步，确定 center
- The centers are moved to seed locations corresponding to the lowest gradient position in a $3 \times 3$ neighborhood.
- 这是为了避免让 seed 落在 edge 上 to reduce the chance of seeding a superpixel with a noisy pixel
怎么 assign pixel？
- 算法的第二步，pixel assignment
- 每个 superpixel 期待的 size 是 $S \times S$
- SLIC 每个 pixel 只寻找 $2S \times 2S$ 内的 center，还是每个 center 寻找 $2S \times 2S$ 内的 pixel
  - 虽然这两个应该本质是一样的，但这个应该关系到对算法的理解吧
- each pixel i is associated with the nearest cluster center whose search region overlaps its location
  - 这也太简单了吧？
  - 这个 nearest 的距离，究竟是空间的距离，还是灰度的距离？
update step —— 调整 cluster center
- adjusts the cluster centers to be the mean $[l\ a\ b\ x\ y]^T$ vector of all the pixels belonging to the cluster
  - 看来这个距离，是灰度和距离的混合体啊
- 10 iterations suffices for most images
postprocessing step
- enforces connectivity by reassigning disjoint pixels to nearby superpixels
整体的算法流程

Distance Measure

因为距离度量同时包含了空间距离和颜色距离，但这两者不在同一空间内，如果 superpixel 大的话，空间距离一下子就会盖过颜色距离，所有有必要 combine the two distances into a single measure，具体的做法是 normalize color proximity and spatial proximity by their respective maximum distances within a cluster, $N_s$ and $N_c$
- $N_s$ 就是 $N_s = S = \sqrt{N/K}$
- fixing $N_c$ to a constant m so that (1) becomes
- simplifies to the distance measure we use in practice:
- Equation (3) can be adapted for grayscale images by setting

Postprocessing

怎么处理 “orphaned” pixels that do not belong to the same connected component as their cluster center may remain
- such pixels are assigned the label of the nearest cluster center using a connected components algorithm.

@article{Achanta2012SLICSC,
  title={SLIC Superpixels Compared to State-of-the-Art Superpixel Methods},
  author={Radhakrishna Achanta and Appu Shaji and Kevin Smith and Aur{\'e}lien Lucchi and Pascal Fua and Sabine S{\"u}sstrunk},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2012},
  volume={34},
  pages={2274-2282}
}

YimianDai/Achanta2012SLICSC.md

2012 - TPAMI - Slic superpixels compared to state-of-the-art superpixel methods

SLIC SUPERPIXELS

Algorithm

Distance Measure

Postprocessing