Image Resizing by Reconstruction from Deep Features

TL;DR

This paper introduces a novel image resizing method that operates in deep feature space, using neural network-based reconstruction to better preserve semantic content and reduce artifacts compared to traditional pixel-space techniques.

Contribution

It proposes a new approach for image resizing by adjusting deep feature maps and reconstructing images, leveraging hierarchical neural network features for improved semantic preservation.

Findings

Reduces artifacts compared to traditional resizing methods

Maintains semantic content and object aspect ratios effectively

Performs well on challenging images and benchmarks

Abstract

Traditional image resizing methods usually work in pixel space and use various saliency measures. The challenge is to adjust the image shape while trying to preserve important content. In this paper we perform image resizing in feature space where the deep layers of a neural network contain rich important semantic information. We directly adjust the image feature maps, extracted from a pre-trained classification network, and reconstruct the resized image using a neural-network based optimization. This novel approach leverages the hierarchical encoding of the network, and in particular, the high-level discriminative power of its deeper layers, that recognizes semantic objects and regions and allows maintaining their aspect ratio. Our use of reconstruction from deep features diminishes the artifacts introduced by image-space resizing operators. We evaluate our method on benchmarks, compare to alternative approaches, and demonstrate its strength on challenging images.