Algorithm: MICF-IFLD

Input: The matrix of all samples X; class labels and concentration information of the training samples. Output: The adaptive matrix, V, in Equation (4), the classifier, f. Begin   Step 1: Construct the concentration features according to MICF.   Step 2: Compute the kernel matrices, Kx and Ky.   Step 3: Obtain orthogonal transformation matrix, W; namely, the eigenvectors of     $K_x\left(-H{K}_{y}H+\mu H\right)K_x$ corresponding to the h largest eigenvalues; Z = WT.;   Step 4: Train the classifier based on the labeled features of the projected samples, Z. Repeat   Step 5: Construct with-class scatter matrix, Sw, and between-class scatter matrix, Sb.   Step 6: Compute transformation matrix, V, according to Equation (4). Until Convergence.   Step 7: Return to the adaptive matrix, V, and classifier, f. End