Context-Aware and Locality-Constrained Coding for Image Categorization

Algorithm 1

Context matching degree calculation.

Input: 𝑥 (feature to be coded), 𝐿 (context length),

𝐁∈𝐑^𝐷×𝐾 (dictionary), 𝐇={𝐻¹,…,𝐻^𝑁𝐶} (relationship matrixes)

Output: 𝑤∈𝑅^1×𝑁𝐶

  {Labeling the input feature  𝑥}

(1)  𝑙_𝑥←argmin_𝑗𝑥−𝐁(∶,𝑗)²,

  {Labeling the feature within the context domain}

(2) Ω(𝑥) ← find the context domain of 𝑥 with size 𝐿

(3)  𝑓={𝑓₁,…,𝑓_𝑁}← find all the features in Ω(𝑥)

(4)  𝑙=𝑧𝑒𝑟𝑜𝑠(1,𝑁)

(5) for  𝑖=1  to  𝑁  do

(6)  𝑙(1,𝑖)←argmin_𝑗𝑓_𝑖−𝐁(∶,𝑗)²

(7) end for

  {calculating the context matching degree}

(8) 𝑤=𝑧𝑒𝑟𝑜𝑠(1,𝑁𝐶),

(9)    for  𝑖=1  to  𝑁𝐶  do

(10)  𝑤(1,𝑖)=∑^𝑁_𝑗=1𝐻^𝑖(𝑙_𝑥,𝑙(1,𝑗))/𝑁

(11)   end for