For timely and precise treatment of certain skin diseases, frequent evaluation of skin barrier function as part of diagnosis has become more necessarily entailed in the clinical environment. To the recent, TEWL (Transepidermal Water Loss) and SCH (Stratum Corneum Hydration) have been widely adopted as non-invasive measurements to simultaneously and supplementarily help assess the function, especially that of patients with epithelial barrier defects. Nevertheless, there are difficulties in obtaining these measurements in practice due to relatively high costs and inaccuracies using different commercial instruments. With advances of visual artificial intelligence, however, this work proposes a CNN(Convolutional Neural Network)-based learning framework called BFE-Net through which TEWL and SCH measurements can be approximately estimated given skin images. Upon assumption that atopic dermatitis (AD) being highly correlated with skin barrier dysfunction, the experimental results on a private dataset for AD show 86.7% accuracy for TEWL and 91.4% for SCH, respectively.