Semantic-Driven Selection of Printer Color Rendering Intents

In this paper we introduce a unified framework that automatically selects the optimal color rendering intent for a given print job. We first present how we extract information from both the image features and the semantic information contained in keywords attached to this image. Then we show how our framework unifies the two inputs to select the optimal ICC rendering intent. The framework is evaluated with a psychophysical experiment on an image data set printed with the ICC media-relative colorimetric and perceptual intents using an Océ large format printer. We find that our method is correctly able to predict the observers preferences in 81% of the images tested when the keyword is included compared to 58% when the keyword is not included. Introduction The aim of our research is to automatically generate optimal print reproductions of images using an inkjet printing system. The International Color Consortium (ICC) provides a consistent workflow to manage the color gamut changes between an original image and its reproduction via a given technology (ink or toner based print, silver halide photograph, electronic display). The ICC has defined four different intents to address the different reproduction objectives a user may have [1]. Each one represents a different color reproduction compromise. In this work we focus on two of these intents, the perceptual and media-relative colorimetric intents. The media-relative colorimetric intent aims for a colorimetric match while the perceptual intent aims for a pleasing reproduction[1, 2]. The visual impact of the media-relative colorimetric intent is likely to cause the reproductions to appear more colorful and with more contrast and the perceptual intent will often prioritize the details over other qualities. From this observation it appears that the selection, by a user looking for an optimal workflow, of one rendering intent versus the others may also depend on the document content. Perhaps for a very colorful image, the user pays more attention to the color reproduction over the details and chooses the rendering intent accordingly. For images with details, the user may choose another rendering intent that produces a better rendering of details. Our aim is to help the user by building a tool to automatically select the optimal rendering intent for a given print job. Much effort has been made towards creating an adaptive processing workflow where the final processing is driven by the input document’s content [3–5]. These adaptive workflows often use a training set of documents which require two inputs: 1. features and 2. performance input. The features used in these workflows help to summarize differences between documents or to group documents into categories. The terms statistics, properties, factors, image characteristics and descriptors have also been used to describe a document’s features. The performance input is often the result of a psychophysical evaluation. Our workflow requires the ability to easily change which ICC profile and rendering intent combination (color workflow) to apply. This requirement excludes the use of time-consuming psychophysical data as the performance input. Instead, we use a set of performance results derived from metric tests, where each metric compares the color workflow performance of a specific perceptible quality attribute [6]. The psychophysical validation results used to test the quality of our first implementation showed that the observers’ preference between color workflows was more significant between rendering intents than between ICC profiles. For most test images, we were able to adaptively select the same rendering intent as the observers when the observers’ preference between rendering intents was significant [7]. Our goal is to improve the automated selection of which rendering intent is optimal for documents that embed several conflicting image characteristics by using semantic information. The inclusion of the semantic information will add an understanding of the scene on a higher semantic level which will improve our prioritization of the conflicting characteristics. The aim of this paper is to introduce a unified framework that automatically selects the optimal rendering intent for a given print job. We first present how we extract information from both the image and the semantic keywords. Then we show how our framework unifies the information to select the rendering intent. We then show some early results obtained with our framework to demonstrate its advantages and compare these framework results with the results of a psychophysical evaluation. A Unified Framework for Image Features and Semantic Information This section explains in two separate subsections, how to extract cues from two very different sources: 1) image pixels features and 2) semantic context. After this we complete the framework by uniting the two methods into a single estimation that can be used for the selection of the best rendering intent. Cues from Numeric Pixel Values We use eight quality attributes, which are Colorimetric Accuracy (CA), Colorfulness (CO), Gamut Boundary (GB), Smoothness (SM), Details (DE), Shadows (SH), Highlights (HL), and Neutrals (NT), respectively [7]. We denote the set of all quality attributes Q. Each quality attribute is represented by 100 expert-selected example images. For a new input image we assess its relatedness to a quality attribute by measuring similarity to its set of example images. This is a typical classification task and we implement a standard method with multivariate Gaussians. 20th Color and Imaging Conference Final Program and Proceedings 323 Mathematical Background Given an example set of images, for each quality attribute, we pre-compute a feature vector for all of the images. The feature vector can contain any type of pixel-based image descriptors, such as lightness, color, or texture features. As a first step we whiten the data by subtracting the mean and dividing through the variance in each dimension separately. The mean and variance are computed globally over all quality attributes q and images. For the rest of this paper the whitened features are used without being referred to as whitened. We estimate for each quality attribute the mean (μq) and covariance (Σq) matrix of its associated point cloud of images in the feature space. Each quality attribute can then be represented as a multivariate Gaussian density distribution in feature space: gq(f) = 1 (2π)|Σq| exp [