*
!
Although the organoleptic method of tea testing has been traditionally used for quality monitoring, an alternative way by machine vision may be advantageous. Although, the three main quality descriptors estimate the overall quality of madetea, viz., strength, briskness and brightness of tea liquor, the exact colour detection in fermenting process leads to a good quality-monitoring tool. The use of digital image processing technique for this purpose is reported to play an effective role towards the production of good quality tea though it is not the only quality determining parameter. In this paper, it has been tried to compare the contribution of the chemical constituents towards the final product with the visual appearance in the processing stage by imaging. The use of machine intelligence supports the process somewhat invariantly in comparison to the human decision and colorimetric approach. The captured images are processed for colour matching with a standard image database using HSI colour model. The application of colour dissimilarity and perceptron learning for the standard images and the test images is ensured. Moreover, the performance of the system is being tried to correlate with the decision made by the organoleptic panel assigned for the tea testing and chemical test results on the final product. However, it should be noted that the optimised result could be achieved only when the other quality parameters such as withering, flavour (aroma) detection, drying status etc. are properly maintained.
Keywords : tea quality, image database, colour dissimilarity, perceptron learning and colour matching etc.
1.
Colour and flavour are the two key parameters for the subjective determination of the overall product quality of many foods processing industries. As technology advances, food-processing industries also try to modernize their processing technology for improving quality of their food products, saving energy and increasing productivity. Quality control and automation in such industry is done by process parameter measurement and monitoring. Colour, size and shape of processed food like sugar, juice, jam, jelly, chips, chocolates, tea etc. are important process parameters determining their quality. Apart from this, texture measurement also leads to a good quality control tool. Tea is an important high value crop through out the world. Since the tea industry occupies a pivotal position in the international trade, sufficient attention has been drawn by the tea industries in regard to quality control of the black tea. Two processes of quality indexing exist so far in tea industry. One is assessment of biochemical parameters and the other one is subjective assessment through tea testing. On the other hand, quality of the final product is based on three main factors, viz., whether hygienically produced, whether free from hazard chemicals (ex. arsenic, cadmium etc.) and whether required constituents are available for cup characteristics. Out of the two processes of quality indexing the subjective tea testing process is highly deserved in most of the tea industry so far reported. The tea tasters obey their own natural language for describing the tea liquor and infused leaf for quality estimation. Some important tea terms are Backey, Body, Bright,
Brisk, Burnt, Colour, Cream, Dry, Dull, Full, Pungent, Strength/Strong, Thin, Coppery, Green and Even etc. These tea terms have a positive correlation with the correctness of different processing stages of tea manufacturing. Besides these, the TF-TR (Theaflavin-Thearubigin) analysis, which is a spectro photometric method is also used with the made-tea for quality judgement. Here total colour and the percentage of brightness of the made-tea are also measured.
!
* surajit@tezu.ernet.in; Tel: +91 3712 267249 / 267045, +91 3715 248197; fax: +91 3712 267005 / 267006.
manab@tezu.ernet.in; Tel: +91 3712 267249/ 231780; EPABX: +91 3712 267007/8/9 Ext. 5252.

Production of black tea involves certain well-defined processing stages governed by many parameters, both natural and controlled. For viewers’ reference, it can be mentioned that the main tea processing stages are withering, rolling
/cutting, fermenting and sorting. Proper control of the each processing stage may stand for good quality of made-tea.
Many tea industries have faced difficulties in producing good quality of tea. Therefore many researchers have attempted to explore techniques to improve quality of the made-tea through process monitoring and control during manufacturing.
The areas in which work have been done for improvement of quality of tea are – monitoring of moisture content of tea leaves, percentage of withering, drying status and relative humidity during withering and substantial improvement have been reported due to adoption of these quality control methods during manufacturing by electronic means
1
. Fermentation is one of the most important processes, which determine the quality of the made-tea. But attempt has not so far been made for implementing quality control in this stage by machine vision. This paper summarizes the aspect of quality control during fermentation by colour matching technique for determination of degree of fermentation where change in colour of rolled tea is a major factor of quality determination. In this work, emphasis has been given in use of machine vision to detect completion of the process when exact colour is achieved in the fermented tea, which in turn leads to good quality final product.
Quality of made-tea depends on almost all the activities involved with the tea manufacturing industry. Beyond that, agronomic practices and transportation of leaf also affect the tea quality
2
. The agronomic practices include the shade, plucking standard, pruning and manure application. Shade improves the quality by filtering the UV ray of sunlight and provides biomass as nutrient for tea bushes. Tea quality is also effected with different standard of plucking such as two leaves and a bud or three leaves and a bud and plucking intervals. The long plucking interval also leads to lowering of tea quality as young shoots and leaves are correlated with good quality tea. Thirdly, pruning can considerably influence the quality of black tea. When tea bushes are pruned, there is a greater effort in the plant to rebuild its structure. Use of different manure also effects in the different pigments present in the tea leafs. Besides them, quality variations can be very much distinguished with different tea clones available. These all effect the constituents of green leaves such as phenolic compounds, caffeine etc. Out of these the caffeine content contributes towards the briskness and creamy property of tea liquor. Here the briskness is a taste sensation and the creaming is the turbidity, which is developed when a good cup of tea is cooled. Again transportation of the tealeaves is a major factors as the tealeaves may get damaged during transportation effecting the different manufacturing processes. During manufacturing, different processing stages affect the quality of the final product. For example, almost 70% (100Kg leaves reduced to 70Kg) withering is desirable to proceed to the next stage, proper rolling and cutting of the tealeaf and finally the optimised fermentation is very important.
Fermentation is one of the most important processes and is one of the most recognizable determining factor for the quality of made-tea. The stage priors to this stage merely condition the tealeaves for reaction to take place during the fermentation. The traditional method of tea fermentation is spreading the rolled and cut tealeaves in a polished floor or trays at a given thickness in a relatively cool ventilated but moist atmosphere. The Continuous Fermenting Machine
(CFM) is also used where the fermentation takes place on a conveyor belt of the machine. Here also the cut tealeaves are spread on the moving belt with uniform thickness and ploughing the leaves from time to time. The optimum temperature is kept at about 27
0
C to 32
0
C, with a hygrometric difference not exceeding 1
0
C so that the relative humidity is to be more then 95%. A series of chemical reactions take place due to severe damage of the leaf cells followed by intermixing of chemicals during previous processes. Although heat, light and pH affect the degradation of carotenoids, which is mostly influenced by oxidized flavonols formed during fermentation. The duration of fermentation varies from 1 hour (for CTC on a warm day) to 3 hours (for orthodox in a cool day). Nevertheless, it is a process of enzymic oxidation of catechins present in damaged and distorted tea leaves resulting in the formation of two groups of coloury compounds, the
Theaflavins (TF) and Therubigins (TR). The TF is golden yellow and TR is reddish-brown in colour. These two compounds together impart to the tea liquor its characteristics colour and taste
3-5
. In addition to the formation of TF and
TR, other chemical changes also take place in the leaf tissues during fermentation process. For instance, proteins get degraded, the chlorophyll are transformed into pheophytins and some volatile compounds are generated due to transformations of certain aroma precursors present in tea leaf. These changes also contribute towards the colour and flavour of made-tea
6-8
. These constituents influence the brightness, briskness, strength, body, and colour of the tea liquor to a great extent with respect to which quality is estimated. Here, the brightness stands for the good-looking tea liquor

implying the absence of any harmful bacteria together with careful manufacture. Briskness is the live taste in the liquor as opposed to flat or soft. The strength implies the presence of body in the mouth. It denotes substance in liquor and is usually combined with a qualifying adjective such as some, a little or good etc. Body is defined by liquor possessing fullness/richness and strength as opposed to thin liquoring teas. Finally the colour is a measure of the depth of the tea colour based on season/growth/grade factors. Out of these quality descriptors, the brightness, briskness and strength are determined by the TF contents whilst body and colour are associated with TR contents. In fact there is a positive correlation between the TF content and market value of black tea. The concentration of TF increases with the progress of fermentation; it reaches a peak value and then starts declining if the fermentation is prolonged. TF gets degraded to TR in this stage. The concentration of TR, on the other hand, goes on increasing with the increase of fermentation time and the body of the liquor becomes thick. The over fermented tea has ‘body’ but lacks other desirable characteristics of good cup of tea. As a whole, the highest possible level of TF content is desired in production of tea. The development of colour and quality during fermentation process is shown in Fig.1. As the advance of the process time both TF and TR contents increase and colour of the tea also gradually changes. At optimum condition when the ratio between TF and TR
Too raw (very little colour and quality
Mellow liquor
(Optimum colour
Too soft (very coloury but little quality)
(Coppery red)
More colour
& less quality
Characteristic
More colour
& more quality
Formation of
TF & TR and quality
Formation of
TR degradation of TF
Desired fermenting time
Fermentation time
Figure 1. Development of colour and quality during tea fermentation. becomes almost 1:10 or 1:9, the colour of the processed tea becomes coppery red. If the process still continues, the degradation of TF to TR occurred and colour becomes more deep imparting less quality tea. So it is seen that the compounds imparting colour and taste to the liquor gradually changes and after the optimum condition they are degraded sharply. Due to this continuously changing phenomenon of the lipid formation, the chemical analysis of the tea during fermentation is somewhat difficult to draw any inference to the quality determination during this process. In view of this, in controlled environment, tea fermentation requires precise monitoring in subjective manner and control of the process as soon as possible as the desired quality is achieved. In such situations, the industries usually need to apply tests for optimization of the fermentation process.
So, accurate detection of the optimum state of the process is the most important phenomenon of the fermentation process for quality perspective in tea processing industry. Once the optimum fermentation is ensured the tealeaves are sent to the dryer to extract the remaining moisture so as to stop fermentation.

The detection of colour and flavour of tea during optimum fermentation condition is a complex problem due to formation of many chemical compounds, which are not directly measurable. In other words, both the parameters represent important means and compute a model for quality control of black tea. Therefore, intensive processing in complex controlled environments of tea fermentation requires supervision (monitoring) and control of the process as early as possible attaining the required colour and flavour. Attempts have been going on developing commercially acceptable quality monitoring tool for tea processing industry. In fact, in the tea trade, the assessment of the quality of the final product by testing is purely subjective matter. Efforts are being made to assess some of the attributes of tea liquor like colour and flavour with the aid of instruments, which is somewhat effective
9-11
. Manual judgment and the colorimetric approach are so far been reported to detect the optimal fermentation time in the tea industry, which are discussed here briefly.
3.1 Manual Method
Optimum colour detection is presently assured by the use of sensory panel of trained personnel’s with eye approximation. In other words, the human experts, since the beginning of the tea industry, have been traditionally detecting the optimum colour during fermentation. This visual inspection is adopted to detect the achieved deep coppery red colour during fermentation at regular intervals. This method is believed as a reliable and faithful subjective assessment of quality, provided exact colour detection is possible. But it has been reported that the exact matching of the colour of tea at optimum fermenting condition is not always possible due to variability of eye approximation of the sensory panel and other factors. Therefore there is every potential of replacing the human visual inspection by machine vision during fermentation process of tea manufacturing. The odour is also judged at the same time to decide the completion of the process, which is to be addressed in a separate study.
3.2 Use of Colorimeter
In some tea factories chemical tests are conducted to find out the correct fermentation of orthodox and CTC tea.
Colorimeter is used in this method. This method is again subjective in nature since the completion of process is again ensured by measuring the intensity of colour i.e., optical density of the ethyl acetate extract made from the fermented tea sample. This correlates the measurement of concentration of TF during the fermentation of tealeaf and to find out optimum fermentation time from the maximum concentration of TF. The procedure of this experiment is - ethyl acetate extract of the fermenting tea sample is collected in a test tube and the intensity of colour is matched by viewing through the column of the liquid against a white back ground. After this the intensity of colour i.e., optical density is measured in a colorimeter at 460 nm against water as blank. This test is repeated at every 10 minutes intervals during fermentation and at 5 minutes intervals when the critical period of fermentation is approached. For CTC manufacture, the maximum depth of colour of the ethyl acetate extract at a particular time corresponds to optimum fermentation. On the other hand, in case of Orthodox manufacture, the process is continued for another 10 minutes after obtaining the maximum depth of colour of the ethyl acetate extract
12
. This method is not easy or results are not always faithful due to a considerable period of time required in pursuing the chemical tests. Moreover, for satisfactory result, it requires high patience and concentration of the concerned person.
Since the experienced tea supervisions depend on visual inspection of the colour variation from green to deep coppery red, this method is established, as the most reliable and faithful subjective assessment of quality provided, exact colour detection is precise. Therefore there is every potential of replacing the human visual inspection by machine vision during fermentation process of tea manufacturing. Digital image processing techniques is being tried to use to evaluate the optimum fermenting condition by matching the colour in a non-destructive way as an alternative for the two above mentioned methods. An attempt is made to describe the method for matching of the colour of the tea at optimum fermenting condition by digital image processing techniques. This section of the paper describes the use of machine vision by imaging the fermenting tealeaves by Charge Coupled Devices (CCD) camera and the effectiveness of colour detection by this method as a quality index of black tea manufacturing. There are two different algorithms so far tested for accurate determination of the exact colour. In the first method the histogram dissimilarity measurement technique is adopted for the purpose and a single stage artificial neural is used in the second method.

The basic categories and hierarchy of rules used by humans in judging similarity and matching of colour patterns are overall colour, directionality & orientation, regularity & placement, colour purity and complexity & heaviness
13
. The overall colour and colour purity are the used categories describing in this paper considering the evenness of colour of the captured images. For representing a colour image, RGB model is most common, where three colour spaces R, G and B represent the image. Hue and saturation of the HSI colour model are used for this purpose of colour matching as the hue represents the impression related to the dominant wavelength of colour stimulus and the saturation expresses the relative colour purity (amount of white light in the pure colour). Not only that, the HSI model is considered for the invariance property for illumination intensity, illumination direction, and viewing direction properties of H and S spaces
14
. Again, the usefulness of this model range from the design of imaging systems for automatically determining the ripeness of fruits and vegetables, to systems for matching colour samples or inspecting the quality of finished colour goods
15
. As the colour information of an image solely resides in the H plane in this model, only the this plane is considered for the matching purpose, the pixel value of which varies from 0 o to 360 o
hue angle corresponding to colour bands Red through
Green to Blue. Since the CCD camera converts the images into RGB model, a conversion operation from RGB to hue
(H) plane has been performed using the equation (1).
H
= cos
−
1
⎡
⎢
⎢
⎢
⎣
( R
1
2
[( R
−
G )
−
G )
2
+
+
( R
( R
−
−
B )]
B )( G
−
B )
⎤
⎥
⎥
⎥
⎦
(1)
If B>G however, H=360 o
- H
4.1 Dissimilarity of image colour scheme
Swain and Ballard in 1991
16
introduced the histogram intersection method as a measure of histogram similarity as well as dissimilarity of images. This technique can be invariably introduced by extracting the colour features of the images to ascertain for the judgement of the colour in the fermented tealeaf images
17
. By definition a colour histogram is a vector where each entry stores the number of hue values of a given coloured image which records the overall colour composition of the image of interest. H space of the HSI model can be related to histogram containing maximum of 360 o and minimum of 0 o
pixel value. Histogram axes are partitioned uniformly with fixed intervals called bins. After plotting the histograms for the hue angles of an input image corresponding histogram comparison for the hue angles is needed for colour matching with a standard image. As the pixel densities in histogram representation are considered in the bin form, vector distances such as Euclidean and L1 (Manhattan) norms are used to quantify the similarity of two colours as numerical vectors represent them. Given a pair of histograms, H(I) and H(Q), of hue of input image I and a standard image Q respectively, each containing n bins, the defined dissimilarity value can be calculated as equation (2).
D { H ( I ), H ( Q )}
= j
N ∑
=
1
| h j
( I )
N
I
XM
I
− h
N
Q j
( Q )
XM
Q
|
=
( N
I
1
XM
I
) j
N ∑
=
1
| h j
( I )
− h j
( Q ) | (2)
Where, N
I
XM
I
is the image size. For a given distance T, two histograms are said to be similar if D<= T, and the images can said to be similar in colour.
Since the fermented tea colour at the optimum condition is deep coppery red, the Hue angles lie near 0 o
in both sides in the Hue angle space (0 o
is fixed for pure red). Therefore, in our application, images of the tea under fermentation fall with pixel value of 0 o
to 120 o
and 290 o to 360 o
in the Hue angle space. Therefore, for simplicity, the pixel values in the range 120 o
< h j
<289 o
are not considered to our images which corresponds to only the colour band green, blue and other adjacent colours only. Before testing by machine vision, one standard image database of optimum fermented tea images is generated supported by the sensory panel decision and colorimeter results. Hue spaces are calculated for each image and stored in a different database, which is used as the final input database. The cross dissimilarity pixel values (DPV)

are calculated among each sample of the hue database and a dissimilarity threshold value (maximum DPV) is calculated.
The results of cross DPV of 10 most standard images are furnished in the Table-I with threshold DPV 1.1992 between
Image 2 and Image 6. Therefore, at the testing phase the dissimilarity between the hue of the input image and these two extreme Images (Image 2 and Image 6) has been calculated. The input image is said to match the target image in colour if both the results are less or equal to the threshold value.
Table-I. Cross-dissimilarity of hue of the database
Images Image1 Image2 Image3 Image4 Image5 Image6 Image7 Image8 Image9 Image10
Image1 0.0000 0.1601 0.2539 1.0163 1.0019 1.0884 1.0081 1.0402 1.0119 0.9215
Image2 0.1601 0.0000 0.4030 1.1612 1.1386 1.1992* 1.1498 1.1852 1.1568 1.0664
Image3 0.2539 0.4030 0.0000 0.8193 0.7841 0.9427 0.7861 0.8402 0.7712 0.6797
Image4 1.0163 1.1612 0.8193 0.0000 0.3955 0.2725 0.3501 0.1395 0.2173 0.2672
Image5 1.0019 1.1386 0.7841 0.3955 0.0000 0.3925 0.1534 0.2965 0.2291 0.1754
Image6 1.0884 1.1992 0.9427 0.2725 0.3925 0.0000 0.2406 0.1568 0.1718 0.3315
Image7 1.0081 1.1498 0.7861 0.3501 0.1534 0.2406 0.0000 0.2273 0.1329 0.1554
Image8 1.0402 1.1852 0.8402 0.1395 0.2965 0.1568 0.2273 0.0000 0.2319 0.0951
Image9 1.0119 1.1568 0.7712 0.2173 0.2291 0.1718 0.1329 0.2319 0.0000 0.1663
Image10 0.9215 1.0664 0.6797 0.2672 0.1754 0.3315 0.1554 0.0951 0.1663 0.0000
* Maximum DPV; Threshold = max DPV.
4.2 Single stage image colour classifier scheme
The use of artificial neural network is one another scheme to discriminate between the target colour required and the different colour in the fermented tealeaf images. A single layer perceptron neural network model is sufficient for this purpose
18
. The database used to train the network consists of images of three categories viz., good fermented (as used in the previous method), low fermented and over fermented tealeaves. The hue spaces are extracted from each of the images of the database and applied to the network as inputs. Again in the input level the entire database is subdivided into two sets, one set consists of images of good fermented tealeaves and the other consists of images of both low and over fermented tealeaves. The decisions about the images to take throughout the training are similar and dissimilar in colour. The first set is considered as the similar colour data set and the rest one is considered as the dissimilar in colour data set. The value of the Perceptron’s decision d(x) is calculated for each of the input hue space and the step activation function is used to discriminate the similar and dissimilar colour images. If h
11
, h
12
, h
13
… h mn are the hue values of an input image and w
11
, w
12
, w
13
… w mn are the weights for the inputs, then the value of d(x) is given by equation (3). d ( x )
= mn ∑ i
=
1 , j
=
1 w ij h ij
(3)
If d(x) > 0, then image
∈
Similar and if d(x) < 0, then image
∈
Dissimilar. Therefore, calculation of the Perceptron’s decision function and using of the step function as the decision-maker are carried out to discriminate at this stage. The randomly chosen weights are adjusted for both categories of images with the following iterative equation (4). w ij+1
=w ij
+
η
[d – y] h ij
(4)
Where, w ij+1
= New adjusted weight
w ij
= Old weight
η
= The learning rate (
≈
0.01)
d = 1, when image
∈
similar
= - 1, when image
∈
Dissimilar
y = 1, when d(x) > 0
= - 1, when d(x) < 0.
h ij
= Hue spaces of the concerned image.

This weight adjustment process is continued until the desired weights are achieved using 100 of different images, where 50 images are of similar colour and rest of the 50 images is of dissimilar colour. When the training is over for whole the database, the final adjusted weight values are stored for discriminate the test images either from similar or dissimilar class of image colour. In the testing phase the same procedure is carried out with the only exception that the output decision is unknown.
5.1 Maximum fermentation time judgement
The colorimetric-results (CR) and its correlation with the machine vision results (DPV and d(x)) of the batch of fermentation process having one of the best results so far is shown in Table-II. The development of required colour in the fermented tea of one batch of manufacturing is furnished in the table. The samples were taken in a CTC manufacturing process conducted in the laboratory of TataTea R&D section, Teok, Assam, India. During fermentation the TF:TR and colour analysis was performed for a lot of 50 kgs of tea which were withered, rolled and fired for the purpose only. The imaging was performed simultaneously and the image matching was performed off line. The time of optimum fermentation therefore can be judged from the peak value achieved by the colorimeter results and the result can be correlated with the machine vision results obtained from the images captured with the same interval of time.
Table-II. Judgement of optimum fermentation time by machine vision
Time (min.) 40 50 60 70 80 85 90 95 100 105 110
CR 0.7 0.8 0.85 0.9 0.9 1.1 1.15 1.1 1.2 1.1 0.9
DPV w.r.t. 2 1.5642 1.6549 2.1369 1.1987 2.1987 2.0091 1.1972 0.9971 0.3789 0.7021 1.2234 d(x) / 10^4 -3.4621 -5.2786 -2.8919 -2.8712 -4.2874 -3.287 -3.981 1.2783 5.7810 2.8971 -3.892
5.2 Calibrated results of the machine vision with TF-TR analysis report
The over all quality of tea is determined by the TF-TR analysis report and the subjective organoleptic methods. The TF-
TR analysis concerns with the percentage of TF content, percentage of TR content, ratio between TF and TR, total colour of the product and the percentage of brightness. The TF content may vary from approximately 0.6% to 1.8% for
CTC and 0.6% to 1% for orthodox depending upon quality of tea. Low quantity of TF content makes the liquor dull and is an indication of inferior quality. Similarly the TR content may very from 8% to 18% (expected) for CTC and 8% to
10% for orthodox with a good quality tea. The TR content gives the tea liquor its depth of colour and more TR content means very strong and coloured liquor with less briskness as caffeine along with TF contribute towards briskness. This method also includes the measurement of total colour, which is the combined contribution of colour from TF and TR present in the tea liquor. Lastly the brightness is directly proportional to the percentage of TF content present. Nine results of the TF-TR analysis in relation to the judgment given by the machine vision are shown in Table-III.
Table-III. Machine vision results vs. TF-TR analysis.
2
Ferm.(min)
100
DPV w.r.t. 2 & 6
& 2.0278e+004
0.9876 & 0.4934 1.1002e+004 1.56 colour
15.67 1:10.04 7.28
16.28 Good
19.76 Very
1.60 good
& 3.3367e+005 13.17 Poor
& 1.60 17.52 1:10.95 8.42 16.54 Good
1.85 22.41 Excellent
&
&
1.87 24.03 Excellent
-1.0234e+005
1.2964e+004
10.60
15.54
Poor
Good
1.672 22.16 Good

The remarks furnished here in the Table-III are not at all well defined but this is the way in which the tea tasters makes the decision about the quality of a given sample of made-tea. They also use the organoleptic tests at this stage where the subjective decisions are made in terms of the strength of liquor, brightness and briskness. The cupped liquor and the infused tea are used at this stage for the organoleptic tests. The proportion of the strength and the briskness should be the almost same which is a sign of good quality tea. In the organoleptic tests the decisions are spelled in terms of excellent, very good, good, fairly good, fair, only fair and poor.
As already mentioned, the quality of the made-tea depends not only on the fermentation process but there are lots of other parameters. Utmost care was taken to maintain the minimum required condition needed for a good quality tea during fermentation. Moreover, withering and drying was also performed with standard parameters. A good correlation between machine vision, chemical analysis and tasters decision was obtained. Out of 30 test samples some good results of 10 samples are furnished in the table where, except two (4 & 10) all conform to TF:TR and organoleptic decision.
Since the aroma is the other most dominant parameter for quality determination during the fermentation process the use of Electronic nose for this purpose parallel to this colour matching approach will be the future work of the paper.
1.
M Bhuyan, “ An Integrated PC based Tea Process Monitoring and Control System ”, Ph.D. Thesis, Gauhati
University, October 1997.
2.
S Sarma, “Biosynthesis of Precursors for Liquor and Flavour Characteristics in Black Tea”, Global Advances in
Tea Science , pp 723-732, Aravali Books International (P) Ltd., New Delhi, 1999.
3.
E A H Roberts, “Tea fermentation”, Economic Importance of Flavonoid Substance , Pergamon Press, New York, pp
468, 1962.
4.
R L Wickremasinghe, “Tea”, Advances Food Research , Vol. 24, pp 229, Academic Press, New York, 1978.
5.
T Yamanishi, “Tea, coffee and cocoa and other beverages”, Flavour research , Marcell Dekker, New York, pp 234,
1981.
6.
P K Mahanta and M Hazarika, “Chlorophyll and Degradation Products in Orthodox and CTC Black Teas and Their
Influence on Shade of Colour and Sensory Quality in Relation to Thearubigins”, J. Sci. Food Agric., Vol. 36 , pp
1133-1139, 1985.
7.
M Hazarika and P K Mahanta, “Some studies on carotenoids and their degradation in black tea manufacture”, J. Sc.
Food Agric, Vol. 34 , pp1390-1396, 1983.
8.
T Takeo and P K Mahanta, “The aroma patterns shown in different kinds of tea”, Int. Tea symposium , Rize, Turkey,
26-28 th
June’ 1987.
9.
M Hazarika, P K Mahanta and T Takeo, “Studies on some volatile flavour constituents in orthodox black tea of various clones and flushes in N. E. India”, J. Sc. Food Agric, Vol. 35 , pp 1201-1207, 1984.
10.
W E Lee, “ A suggested instrumental technique for studying dynamic flavour release from food products”, J. Food
Sc., Vol. 51 , pp 249-250.
11.
T Takeo and P K Mahanta, “ Comparison of black tea aromas of orthodox and CTC tea and black teas made from different varieties”, J. Sci. Food Agric, Vol. 34 , pp 307-310, 1983.
12.
M R Ullah, K L Bajaj & S Chakraborty, “Fermentation test by Dr. M R Ullah”, Factory Floor Fermentation Test , pp 3-6, Tocklai Experimental Station, Jorhat, India, 1979.
13.
A Mojsilovic, J Kovacevic, D Kall, R J Safranek and K Ganapathy, “The vocabulary and grammar of color
Patterns”, IEEE Trans. Image Processing, Vol. 9 , No. 3, pp 417-431, March’2000.
14.
T Gevers and AWM Smeuders, “Color based Object Recognition”, Pattern Recognition, Vol. 32 , pp 453-464, 1999.
15.
Gonzalez R C Gonzalez & R E Woods, Digital Image Processing , pp 229, Addition-Wesley Publishing Company,
1993.
16.
M. J. Swain and D. H Ballard, “Color Indexing”, Int. Journal of Computer Vision, Vol. 7(1) , pp 11–32, 1991.
17.
S Borah & M Bhuyan, “Non-destructive testing of tea fermentation using image processing”, Proc. NDT 2002 , pp
199-204, BINDT, 17-19 th
Sept.’ 2002.
18.
S Borah, M Bhuyan & H Saikia, “ANN based Colour Detection in Tea Fermentation”, Proc. of ICVGIP’2002 , pp
226-229, ISRO, Ahmedabad, India, 16-18 th
Dec.’ 2002.