Object Color Recognit Ion Using CNN

Object color recognition is a sub-domain of computer vision, which aims to find the major color of the object on a given image. This task actually contains two main aspects: making the computer understand color names in human world, as well as recognizing the most salient color of foreground of an image. However, existing methods tend to solve only one aspect. This paper proposes a solution based on deep learning, which could handle the color name learning and major color recognizing as a whole. By introducing the convolutional neural network, our solution could learn color features from the raw image, which simplifies the setup and training procedure of the model, as well as learning the potential association of inputs. This helps to achieve a better recognition result. Since the designed CNN model is deep, we also adopt some optimizers like the Inception module, which effectively prevents the system from computational blow up and over fitting.Our experiments show that the proposed solution achieves a better accuracy and ro-bustness over traditional methods in item category specified color recognition.