Inverse image editing

Abstract

Applicationsre-edited image Figure 1: Given a source image and an edited copy (left), our system automatically recovers a semantic editing history (middle), which can be used for various applications, such as re-editing (right). In this case, the second editing step of the recovered history, involving hue modification, is altered to change the berries to a difference color. Image courtesy of Andrea Lein. AbstractWe study the problem of inverse image editing, which recovers a semantically-meaningful editing history from a source image and an edited copy. Our approach supports a wide range of commonlyused editing operations such as cropping, object insertion and removal, linear and non-linear color transformations, and spatiallyvarying adjustment brushes. Given an input image pair, we first apply a dense correspondence method between them to match edited image regions with their sources. For each edited region, we determine geometric and semantic appearance operations that have been applied. Finally, we compute an optimal editing path from the region-level editing operations, based on predefined semantic constraints. The recovered history can be used in various applications such as image re-editing, edit transfer, and image revision control. A user study suggests that the editing histories generated from our system are semantically comparable to the ones generated by artists.

Multi‐image matching for object recognition

“…However, multiple segmentations are generated in [10], and the scale‐invariant feature transform (SIFT) features cannot be extracted from these segmentations. Hu et al [11] introduced PatchNets, a compact, hierarchical representation that describes structural and appearance characteristics of image regions, for use in image editing. A low‐dimensional representation of an image which preserves its inherent information from the original image space was learned in [12]; however, this approach cannot find the object region and be combined within the bag of words (BoW) framework.…”

Efficient image-aware version control systems using GPU

“…It is important to note that filter operations cannot be defined by a bijective function and, for this reason, are not considered here: for example, applying a blur filter would generate a delta that consists of all the pixels that are not black. Hu et al [30] can detect this kind of transformation as well as many others, but their approach requires at least 3 min to process an image of 512 512 pixels and so is not feasible for use in version control. When looking at both the original (A) and the modified (B) images in Figure 9, it is easy to see that an 180 ı clockwise rotation has been performed.…”

“…Hu et al [30] introduced an approach to recover a semantically meaningful editing history from a source image and an edited one. Their technique supports the detection of global and local linear and nonlinear color changes, the insertion and removal of objects, and cropping.…”