With the fourth industrial revolution, the focus on innovative uses of technology in the printing and mass customization space has become unparalleled. Mass customization that clubs together the flexibility and personalization of custom made business manufacturing is growing at a very high scale and pace and is becoming an important business concept. Plenty of brands are adopting it today to get better recognition in their service and product lines.
In recent years, organisations have invested heavily in artificial intelligence (AI), specifically machine learning techniques, to help make the document processing and pre-press flows more efficient and introduce new capabilities which were not feasible earlier. Many of these techniques are focused on analyzing and manipulating image data in particular.
Crop mark identification:Sophisticated customers of printed material often include marks called “printer’s marks,” “registration marks,” or “crop marks” on their design. These are standard in the printing industry and are indications or instructions to allow manual verification of alignment for cutting and folding. In the mass customization business, however, the manufacturing process is separated from the design process and highly automated. These markings in many cases can interfere with, rather than help, in producing quality results. As a result, attempts have been made to detect these marks through automated processes and adjust designs accordingly to match the intent of the customers. This is highly efficient compared to the alternative of manually verifying and correcting all designs.
Enhancing low resolution images: Customers often do not have access to high resolution or vector images, especially when designing large items, such as banners. Simply scaling the customer’s provided imagery using a standard upscaling algorithm will leave rough edges on lines and insert other undesirable artifacts in the images. It has been found that an appropriately designed neural network can compensate for this by analyzing the image and removing “noise” from it while upscaling, all the while preserving the texture and fine details of the image. The quality of the results is significantly better than most commonly used techniques in the graphics industry. This vastly increases the range of customer imagery that we can safely accept for printing or other forms of decoration.
Removing the background of images:Isolating the principal subject matter of an image and removing the rest – the background – is not straightforward, especially when the image is complex to begin with, and has backgrounds that are not simple solid colors. For instance, reliably isolating a person’s face to provide a portrait image from a snapshot of the person in an urban setting or in a landscape is difficult to do in an automated fashion, and time-consuming and tedious to do manually. There has been an increasing usage of neural networks trained on a large set of sample imagery, to detect and isolate the salient portions of images accurately in order to preserve them while removing the rest of the image content, obviating the need for manual manipulation of such imagery.
Text legibility:A number of factors can contribute to the illegibility of text on a design. The customers are often unable to tell whether a design that looks legible on screen will have problems with the text as it appears when the design is printed, engraved, embroidered, etc. Problems range from text fonts with thin strokes, lack of contrast between the text and the background or busy backgrounds, and improper ink color specifications for text, to text that is simply too small in size for the desired product.
Research has yielded neural networks that can handle such cases reliably. These can analyze the customer’s imagery to isolate and identify all fragments of text and process the rest of the image as well, applying various heuristics for legibility, and then attempt to fix the problems. In case the problem is too difficult for automated adjustment, the design can be redirected to a professional designer for fixing. In this way, we keep throughput high for automated processing, reserving professional designers’ services for only the most challenging cases.
As can be seen from the above examples, the printing and mass customization industry strives to maintain the highest standards of quality in reproducing a customer’s design for any customized product, while maximizing the use of automation and reducing the need for manual intervention. Our research into using AI techniques for many common problems seen in our field has seen many successes so far, and we expect to see many more as we continue to invest in this area.
(The article is authored by Satish Pai, GM, Artwork Technology, Cimpress & David Greenberg, Director of Technology, Cimpress and the views expressed in this article are their own)