Ayon Tarafdar*,  Ranjna Sirohi**,  Anurag Kushwaha***

Nature has imparted multitude of colours to different food commodities which helps to distinguish one product from another. Due to this natural phenomenon, mere visual inspection allows us to separate superior quality fruits, vegetables, and, dairy and meat products, both in raw and cooked states, from that of defective, spoiled or sub-standard products. While inspecting food products manually, a human inspector can assign a grade (high or low) to individual food item based on its appearance (or even weight). This process is tedious, time consuming and subjective. The major challenge is to mimic this biological ability via industrial processing machineries through advanced engineering interventions. Scientists and other researchers worldwide have strived relentlessly to establish the connection between human vision and machines which led to a new scientific field popularly known as machine vision. Colour sorting using machine vision is quite popular in food industries and allows food processors to classify and store products based on their colour attributes in a cost-effective, precise, non-destructive and efficient way. For the food sector, colour sorting serves the purpose of risk management, ease of product handling and inspection, reducing drudgery and faster food processing.

Industries employ different methods to analyze the colour of food products. The two most commonly used techniques involve colourimetry and spectrophotometry. Colourimetry is based on the visualization of the primary colours of viz., red, green, blue (RGB). However, it is unable to identify the secondary and tertiary colours (cyan, magenta, yellow, black or CMYK) which make its utility limited. Spectrophotometry involves the detection of colours over the entire visible light wavelength range of 400−700 nm that helps standardize colours with high reliability but at the cost of complex hardware requirements. Taking into consideration the limitations and advantages of the two technologies, various developments have been made to reach a middle ground wherein precision colour measurements of food products can be done in a cheaper way. Therefore, cost effective systems were envisioned that employed a new colour model called the L*a*b* model which encompassed both the RGB and CMYK colours. The L*a*b* model is now internationally recognized for colour measurement and was developed in 1976 by the Commission Internationale d’Eclairage (CIE) also called the International commission on illumination. Here L* ranges from 0 to 100 corresponding to darkness and whiteness, respectively. The a* and b* values both range from −120 to +120. The a* component represents redness (negative) to greenness (positive) while the b* component represents blueness (negative) to yellowness (positive). All food products show colours as a function of L*, a* and b* enabling its easy identification and classification which is crucial to sorting. Before moving on to discussing the simplest form of colour sorting equipment, it is important to note that the earlier colour models, RGB and CMYK are device dependent whereas L*a*b* is a standardized colour system.

Setting up of a system for colour sorting using machine vision could constitute a simple digital camera (2 megapixel or above), standard illumination and a computer software for image processing (Fig. 1). Standard illuminations typical to food research includes type A (2856 K), C (6774 K), D65 (6500 K) and D (7500 K). Of these standard light sources, C, D65 and D represent various daylight intensities. The lighting should be adjusted at an angle of 45° from the camera lens axis to ensure uniform illumination and minimize harsh shadows. In an industrial setting, food would move on a conveyor belt and real-time images would be sent to a computer where the image will be processed to identify the status of the food product (good or bad).

Figure 1. A simple colour sorting system for food products

However, it is important to note that the image obtained from a digital camera could be in RGB format which needs to be converted to the standard L*a*b* form for analysis. In such a case, various models and equations are available which can be used but the discussion of these is beyond the scope and purpose of this article. There are also image processing software which have an inbuilt colour conversion system of RGBà Lab (not standardized) which is further converted to L*a*b* (standardized) again using simple equations. An example of such as case has been illustrated in figure 2 where images of button mushroom samples were analyzed using Adobe Photoshop and converted directly to Lab.

Figure 2. Example of conversion of RGB to Lab colour system using image processing software

In general, the software of a colour sorting system based on machine vision would execute a series of steps that include: (i) image acquisition, (ii) image processing, (iii) feature extraction and, (iv) decision and control. The first two terms are relatively simple and need no explanation whereas the last two deal with extracting typical features of the image (noise removal, focussing on the product etc.) and subjects it to a decision support system. Based on the colour attributes collected, the decision support system gives the final verdict on the product class as per information available to it in a pre-designed database.

A usual disadvantage associated with this kind of system involves the incomplete visualization of a food surface due to lack of product rotation. This is particularly a problem where colour based systems are used for defect detection. In such a condition, additionally photosensors and microcontrollers are used along with limit switches to determine the position of the product. Rollers on conveyer systems can be also given to move the food product in different positions (Fig. 3). Use of multiple cameras has also been shown to give promising results. This kind of method has been proven to be very effective. For instance, the colour grading of strawberries using this technique was found to be 88.8% effective. It is also used for sorting of apples, oranges and tomatoes.

Figure 3. Change in orientation of apples using mechanical systems for defect identification (dark spots/damage)

In conclusion, semi- or fully automatic systems for colour sorting of food products based on machine vision system is a breakthrough in the food sector. Care must be taken in deciding lighting conditions, choice of colour sensors, object to camera distance and adequate image rotation to acquire precise product information. The development of cost effective and improved hardware and software assemblies are underway. Probably in the near future we will witness a revolution bringing more sophisticated artificially intelligent colour sorting systems.

*Division of Livestock Production and Management, ICAR-Indian Veterinary Research Institute, Izzatnagar, Bareilly 243 122, Uttar Pradesh, India

**Department of Chemical and Biological Engineering, Korea University, 145, Anam-ro, Seoungbuk-gu, Seoul 02841, Republic of Korea 

***Division of Agricultural Engineering, ICAR-Indian Agricultural Research Institute, Pusa 110 012, New Delhi, India


Yam, K. L., & Papadakis, S. E. (2004). A simple digital imaging method for measuring and analyzing color of food surfaces. Journal of food engineering61(1), 137-142.

Tarafdar, A., Shahi, N. C., & Singh, A. (2020). Color assessment of freeze‐dried mushrooms using Photoshop and optimization with genetic algorithm. Journal of Food Process Engineering43(1), e12920.

Eissa, A. A., Khalik, A. A., & Abdel, A. A. (2012). Understanding color image processing by machine vision for biological materials. Structure and Function of Food Engineering, 227-274.