Colour image analysis as a non-invasive method in assessing fresh fish meat content.

Teusdea, Alin Cristian ; Mintas, Olimpia Smaranda ; Bals, Cristina Adriana 等

1. INTRODUCTION

The major aim of this paper is to find a non-invasive and fast research method to assess the quality of fish meat from quantitative and hygienic point of view (Fumiere et et al., 2004). This method must be proper for industrial application and for aggressive environments. Thus, the main goal is to identify (i.e. to classify) the fish meat content in percentages for the five classes: red meat, fat, muscle, bone, skin (Tan et al., 2000).

For assessing the content of the fresh fish meat many colour image analysis researches were done worldwide (Del Moral et al., 2007; Hopkins, D.L et al., 2004; Pipek et al., 2004). There were made classifications for only three classes (muscle, bone, fat).

In this paper the classification process is based on converting the digital scanned fresh fish meat images in CIE [L.sup.*][a.sup.*][b.sup.*] colour space which has the advantage of being a linear chromatic space. Furthermore it has a training stage to asses the chromatic limits for each class by using the reference class images.

For an accurate analysis there is done a comparison ("calibration") between the fresh fish meat content results of the NIR-CIA method with the classical gravimetric invasive method one. This is accomplished with Fourier correlation analysis of the meat content "profiles".

2. BACKGROUND

2.1 CIE [L.sup.*][a.sup.*][b.sup.*] COLOUR SPACE

Digital images are usually coded in RGB (Red Green Blue coordinates) colour space. Each coordinate has the integer number range between 0 and 255. This colour space is used to capture and transfer the digital images.

In many industries the CIE [L.sup.*][a.sup.*][b.sup.*] colour space is used because it is approximately a uniform space. The [L.sup.*] coordinate denotes the lightness and has the range between 0 and 100. The zero value for [L.sup.*] represents black and the 100 value represents white. Coordinates [a.sup.*] and [b.sup.*] have no specific numerical limits, and they denote the chromaticity. Positive [a.sup.*] values denote red and negative [a.sup.*] values denote green. Positive [b.sup.*] values denote yellow and negative [b.sup.*] values denote blue. The chromatic plane is organized so that the [a.sup.*] and [b.sup.*] axes have complementary colours at their ends.

The main issue, when one wants to process digital captured images, is to transform their RGB coding to CIE [L.sup.*][a.sup.*][b.sup.*] coding. This transformation is done in two steps. First the RGB space is transformed in XYZ uniform space and next in CIE [L.sup.*][a.sup.*][b.sup.*] space. The RGB to CIE [L.sup.*][a.sup.*][b.sup.*] transformation numerical algorithm is as follow (Leon et al., 2005)

r = R255; g = g/255; b = B/255 (1)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)

where

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

According to CIE [L.sup.*][a.sup.*][b.sup.*] 1992 model the constants are a = 0.055 and [gamma] = 2.2 (Leon et al., 2005). The tristimulus point coordinates, ([x.sub.0]; [Y.sub.0]; [Z.sub.0]), depend on the illuminating source used to capture the image.

The images coded in CIE [L.sup.*][a.sup.*][b.sup.*] colour space, described above, can be compared more accurately.

2.2 Fourier correlation

The correlation process may be a statistical analysis or a Fourier spectral one. The normalized Fourier correlation coefficient, NFCC , can be built from the Fourier analysis, described (Grierson, 2006) by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)

where f(x), g(x) are two functions, F(u), G(v) are their Fourier transforms (Pytharouli & Stiros, 2005; Pytharouli & Stiros, 2008).

The statistical significance of the correlation coefficient values is: 0.10-0.29 for weak; 0.30-0.49 for average, 0.50-1.00 for strong.

3. RESULTS AND DISCUSSIONS

Two carp chops and two carp fillets were scanned with a Genius Vivid Colour flatbed scanner (with F11 standard illumination lamp). The optical resolution of the scanned images is 300 x 300 dpi (i.e. 25.4 mm/300 = 0.0846 mm x 0.0846 mm of a pixel) and the background is black. The carp chops were scanned on both sides, thus there were 6 scanned images. The carp chops and fillets were scanned in fresh conditions.

The aim of the image colour analysis, of fresh carp meat, is to asses an algorithm which discriminates the targeted classes: red meat, fat, muscle and if it is possible the bone and skin.

CIE [L.sup.*][a.sup.*][b.sup.*] colour space coding is used by the numerical discrimination algorithm. So, the first step is to calculate the CIE [L.sup.*][a.sup.*][b.sup.*] coordinates of the scanned images.

The second step is the discrimination process trainning for the targeted classes. In this step the operator selects regions of targeted classes and determines their CIE [L.sup.*][a.sup.*][b.sup.*] properties (i.e. numerical limits).

The CIE [L.sup.*][a.sup.*][b.sup.*] limits of the targeted classes can be represented as rectangular prisms (or boxes) that embed the CIE [L.sup.*][a.sup.*][b.sup.*] points cloud of the carp chops and fillets (figure 2).

Third step consists in classification of the original scanned carp meat image into targeted classes.

Counting the number of the carp chop pixels that belongs to the targeted classes and generating the meat content report is the last step.

Classical analysis (gravimetric invasive method) was performed on 6 samples of fresh carp meat.

The skin class is present in carp chops images in a relevant way. In the carp fillets images the skin class is not accurately discriminate, because the skin is located on the other side of the fillets and it was not scanned.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

In order to compare the accuracy of the NIR-CIA (Non-Invasive Research--Colour Image Analysis) method classical content analysis was done.

4. CONCLUSIONS

In the present paperwork was assessed the percentage content of fresh fish meat (i.e. carp chops and fillets). This task was done using the non-invasive research with colour image analysis technique (NIR-CIA). The percentage content of the fresh fish meat was generated as a result.

The Fourier correlations of the fish meat content "profiles" generated by NIR-CIA and classical gravimetric analysis are presented in figure 3. In this way assessing the fish meat content has higher statistical accuracy.

As previously mentioned, the correlations are strong and the NFCC values are close.

The strong correlation results for the fresh carp meat denote that the NIR-CIA method is an accurate method in assessing the fresh fish meat content and the quality of fish products.

This method prove to be non-invasive and very fast for revealing quality defects, including red meat and quality parameters such as protein content (muscle), fat content, in fresh carp meat.

5. REFERENCES

Del Moral, F.G.; O'Valle, F.; Masseroli, M. & Del Moral, R.G. (2007). Image analysis for automatic quantification of intramuscular connective tissue in meat, Journal of Food Engineering, Vol. 81, 33-41, ISSN 0260-8774.

Fumiere, O.; Veys, P.; Boix, A.; von Holst, C;. Baeten, V. & G. Berben (2004). Methods of detection, species identification and quantification of processed animal proteins in feedingstuffs, Biotechnol. Agron. Soc. Environ., Vol. 13 (S), 59-70, ISSN 1370-6233.

Grierson, B. A. (2006) FFT's, Ensembles and Correlations, Available from: http://www.ap.columbia.edu/ctx/ctx.html, Accessed: 2007-08-12.

Hopkins, D.L.; Safari, E.; Thompson, J.M & Smith, C.R. (2004). Video image analysis in the Australian meat industry--precision and accuracy of predicting lean meat yield in lamb carcasses, Meat Science, Vol. 67, 269-274, ISSN 0309-1740.

Leon, K.; Mery, D.; Pedreschi, F. & Leon, J. (2005). Color Measurement in [L.sup.*][a.sup.*][b.sup.*] units from RGB digital images, Food Research International, Vol. 39, Issue 10, (December 2006), 1084-1091, ISSN 0963-9969.

Pipek, P.; Jelenikova, J. & L. Sarnovsky (2004). The use of video image analysis for fat content estimation, Czech J. Anim. Sci., Vol. 49, No. (3), 115-120, ISSN 1212-1819.