1. Introduction
Digital images can be different in terms of color tones, contrast, clarity, vignetting, and etc. Here we refer such characteristics of digital images as picture styles. With the popularity of photo editing and sharing services such as Instagram, Facebook and Flickr that are available on mobile devices, many digital images generated by users nowadays are captured by a wide range of devices (e.g., smart phones and digital slrs) and processed using different photography effect filters (e.g., “lomo-fi” and “lord-kelvin” available in Instagram) to get distinct picture styles with strong personal artistic expressions. Recall that the goal of object recognition is to recognize natural scenes [14], daily objects [5], or fine-grained species [18], [23] based on digital images, it is natural to extend the scope of object recognition from standard laboratory images to photos in the wild for daily use. Although there are a large number of picture styles, their contributing factors can be separated into 3 major categories: (1) scene radiance, (2) image pipeline, and (3) post processing. In Fig. 1, we show three pairs of images of the same objects to illustrate different picture styles. We show 3 pairs of images about the same objects with different picture styles. The differences between (a) and (b) are mainly due to different scene radiances (illumination condition). (c) and (d) are of the same object and taken under the same condition by a digital SLR and a webcam respectively, representing two different image pipelines. (f) is an image obtained by applying Instaram™ lomo-fi effect filter as a post-processing step to image (e), representing one specific photography effect. In the upper left is an original image from the Oxford Flower data set. In the lower left is the lomo-fi version of the image. We select two regions at the same location from the two images (indicated by red boxes), and show the pixel patch, gradient, and SIFT descriptor for each of them. We then plot the difference between two descriptors in the right.