Low-level image features are transformed into abstract representation of objects through the hierarchical processing in the visual cortex. An intermediate-level cortical area V4 has been considered to play a crucial role in the construction of “objects”. Specifically, V4 neurons have been reported to code rich contour information that can be the basis for the intermediate representation of objects. We hypothesize that the encoding of local contour features is essential for the intermediate representation of objects, and therefore, the characteristics of the cortical contour-coding such as simultaneous coding and population-based representation also emerges automatically in the internal representation of artificial hierarchical networks. We compared the representations of contour features in V4 and the mid-layer of deep convolutional neural network (DCNN) pre-trained for object classification. To enable the direct comparison of the two representations, we used the encoding model which translated the activities of the mid-layer of DCNN to the neural responses in V4. We found the strikingly similar characteristics between the neurons and models in the responses to contour features such as curvature, closure, and symmetry and their simultaneous representation. The population response patterns also showed the significant similarity. These results indicate that the mid-layer of DCNN capture the contour representation comparable to V4 neurons including simultaneous coding of multiple features and population-based representation, suggesting the fundamental roles of local contours and the cortical schemes in the intermediate-level representations.