KmsGC: An Unsupervised Color Image Segmentation Algorithm Based on 𝐾-Means Clustering and Graph Cut

Algorithm 1

The basic structure of the presented algorithm KmsGC.

1st  stage. Clustering

    Input: image 𝐼

    Output: initial pixel partition.

    (1) initialize 𝐾_max

    (2) for 𝑖=2: 𝐾_max

    (3)  𝑘-means(𝑤,2)  ⧵⧵  𝑤 is the cluster to be partitioned into 2

    (4)  calculate intra, inter, CP(𝑖)

    (5)  calculate 𝜎_𝑖

    (6)  𝑤 = find(max(𝜎_𝑖))

    (7) endfor

    (8) 𝑘 = find(min(CP(𝑖)))

    (9) Return the pixel partition corresponding to 𝑘.

2nd  stage. Image segmentation

    Input: initial pixel partition

    Output: segmented image

    (1) Initialization.

    (2) Determine the data term as (7)

    (3) Determine the smooth term as (8)

    (4) Construct a graph correspond to the energy function

    (5) Compute the minimum cost multi-way cut using the expansion algorithm.

    (6) Return segmented image