Image Aesthetics Assessment Using Multi Channel Convolutional Neural Networks

Abstract

Image Aesthetics Assessment is one of the emerging domains in research. The domain deals with classification of images into categories depending on the basis of how pleasant they are for the users to watch. In this article, the focus is on categorizing the images in high quality and low quality image. Deep convolutional neural networks are used to classify the images. Instead of using just the raw image as input, different crops and saliency maps of the images are also used, as input to the proposed multi channel CNN architecture. The experiments reported on widely used AVA database show improvement in the aesthetic assessment performance over existing approaches.

Image Aesthetic Assessment: A Deep Learning Approach Using Class Activation Map

“…These models are modified and used as fine-tuning models of pre-trained networks. However, as stated in [4,9], using only raw images to train these architectures are not suffice for accurate aesthetic class assessment. Hence, various kinds of pre-processed images are fed to these models for feature extraction.…”

“…2, three different deep CNN models, VGG19, InceptionV3 and ResNet50, have been used in the proposed two-channel architecture. It was reported in [9] that using a three-channel network with one channel taking random crop images as input helps to improve the IAA results. In order to have a complete set of comparisons, the proposed two-channel network is also compared with a three-channel network model with three variants of the random cropped image.…”

“…Table 3 shows an accuracy comparison of existing approaches in the literature on the AVA dataset. From the table, it is shown that Single column network accuracy with VGG19 architecture has improved the results as compared to SCNN introduced in [4] and single-column network architecture with the base model of VGG19 used in [9]. [9] has used two variants of saliency map images as one of the inputs in multi-channel CNN.…”

“…In this paper, CAM data is used for image aesthetic task. A single-channel network with VGG19 as a base model shows that CAM features turn out to be better in aesthetics assessment task than saliency map images used in [9]. Lu et al [4] also introduced DCNN in the same paper, which has improved the accuracy of 73.25%.…”

“…These parallel layers represent different scene categories and are fed to fully connected layers for binary classification. Doshi et al [9] proposed three-column CNN based approach, where pre-trained VGG19 architecture was used. They have used two variants of saliency map images as one of the inputs in multi-channel CNN and reported improved IAA performance.…”