Using grayscale images for object recognition with convolutional-recursive neural network

There is a common tendency in object recognition research to accumulate large volumes of image features to improve performance. However, whether using more information contributes to higher accuracy is still controversial given the increased computational cost. This work investigates the performance of grayscale images compared to RGB counterparts for visual object classification. A comparison between object recognition based on RGB images and RGB images converted to grayscale was conducted using a cascaded CNN-RNN neural network structure, and compared with other types of commonly used classifiers such as Random Forest, SVM and SP-HMP. Experimental results showed that classification with grayscale images resulted in higher accuracy classification than with RGB images across the different types of classifiers. Results also demonstrated that utilizing a small receptive field CNN and edgy feature selection on grayscale images can result in higher classification accuracy with the advantage of reduced computational cost.