Cheng-Jin Du

Prediction of beef eating qualities from colour, marbling and wavelet surface texture features using homogenous carcass treatment

Colour, marbling and surface texture properties of beef longissimus dorsi muscle are used in some countries to grade carcasses according to their expected eating quality. Handheld VIA systems are being used to augment the grader assessments, however attempts have been made to develop higher resolution image systems to give consistent and objective predictions of quality based on these properties. Previous efforts have been unable to model sufficiently the variation in eating quality. A new approach has been applied whereby beef carcasses were subjected to homogenous post-slaughter treatment to minimize variation in eating quality related to other factors such as chilling temperature and hanging method. Furthermore a wider range of features were used to better characterize colour and marbling and the wavelet transform was used to characterize texture. Objective and sensory panel tests were performed to evaluate the beef eating qualities. Classical statistical methods of multilinear regression (MLR) and partial least squares regression (PLSR) were used to develop predictive models. It was possible to explain a greater portion of variation in eating quality than before (up to r^2=0.83). Carcasses were classified as high or low quality with a high rate of correct classifications (90%). Genetic algorithms were used to select the model subsets.

Prediction of beef eating quality from colour, marbling and wavelet texture features.

Beef longissimus dorsi colour, marbling fat and surface texture are long established properties that are used in some countries by expert graders to classify beef carcasses, with subjective and inconsistent decision. As a computer vision system can deliver objective and consistent decisions rapidly and is capable of handling a greater variety of image features, attempts have been made to develop computerised predictions of eating quality based on these and other properties but have failed to adequately model the variation in eating quality. Therefore, in this study, examination of the ribeye at high magnification and consideration of a broad range of colour and marbling fat features was used to attempt to provide better information on beef eating quality. Wavelets were used to describe the image texture of the beef surface at high magnification rather than classical methods such as run lengths, difference histograms and co-occurrence matrices. Sensory panel and Instron analyses were performed on duplicate steaks to measure the quality of the beef. Using the classical statistical method of partial least squares regression (PLSR) it was possible to model a very high proportion of the variation in eating quality (r(2)=0.88 for sensory overall acceptability and r(2)=0.85 for 7-day WBS). Addition of non-linear texture terms to the models gave some improvements.

Automatic measurement of pores and porosity in pork ham and their correlations with processing time, water content and texture

Pores formed in pork ham have a significant effect on its quality. However, they are mostly characterised using manual methods with special devices. In this paper, an automatic method for pore characterisation of pork ham was developed using computer vision. To segment pores from images of pork ham, three stages of image processing algorithm were developed, i.e., ham extraction, image enhancement, and pore segmentation. From the segmented pores, the porosity, number of pores, pore size, and size distribution were measured. The statistical analysis showed that 79.81% of pores have area sizes between 6.73×10(-3) and 2.02×10(-1)mm(2). Furthermore, it was found that the total number of pore (TNP) and porosity highly negatively related to the water content of pork ham (P<0.05), and had negative correlations with the cooking and cooling time. However, for texture analysis, positive correlations were found between the pore characterisations and WBS, hardness, cohesion, and chewiness, respectively, while springiness and gumminess were negatively related to TNP and porosity.

Segmentation of complex food images by stick growing and merging algorithm

An algorithm for segmenting images using stick growing and merging is described, which consists of four major steps: stick initialisation, stick merging, subregion merging, and boundary modification. It is started from an initial decomposition of the image into small sticks and non-sticks. The small sticks are merged to obtain the initial subregions on the basis of homogeneity criteria. Then smaller subregions with only one stick are merged into larger subregions and subsequently all subregions are merged into regions according to the criteria. Finally, non-sticks and separate small sticks are merged and the degree of boundary roughness is reduced by boundary modification. The algorithm was successfully used to segment many types of complex food images including pizza, apple, pork and potato.

Development of a hybrid image processing algorithm for automatic evaluation of intramuscular fat content in beef M. longissimus dorsi.

An automatic method for estimating the content of intramuscular fat (IMF) in beef M. longissimus dorsi (LD) was developed using a sequence of image processing algorithm. To extract IMF particles within the LD muscle from structural features of intermuscular fat surrounding the muscle, three steps of image processing algorithm were developed, i.e. bilateral filter for noise removal, kernel fuzzy c-means clustering (KFCM) for segmentation, and vector confidence connected and flood fill for IMF extraction. The technique of bilateral filtering was firstly applied to reduce the noise and enhance the contrast of the beef image. KFCM was then used to segment the filtered beef image into lean, fat, and background. The IMF was finally extracted from the original beef image by using the techniques of vector confidence connected and flood filling. The performance of the algorithm developed was verified by correlation analysis between the IMF characteristics and the percentage of chemically extractable IMF content (P<0.05). Five IMF features are very significantly correlated with the fat content (P<0.001), including count densities of middle (CDMiddle) and large (CDLarge) fat particles, area densities of middle and large fat particles, and total fat area per unit LD area. The highest coefficient is 0.852 for CDLarge.

CORRELATING IMAGE TEXTURE FEATURES EXTRACTED BY FIVE DIFFERENT METHODS WITH THE TENDERNESS OF COOKED PORK HAM: A FEASIBILITY STUDY

Being one of the most important attributes that affect the eating quality of meat products, tenderness is still mostly evaluated using sensory panel and instrumental methods. It is desirable to develop a fast, non-destructive, accurate, and on-line technique for tenderness evaluation of meat products. As an objective, consistent, rapid, and automatic technique, computer vision could be employed to complete such a task. The relationships between tenderness and image texture features of pork ham were investigated in this study. Fifty observations were made, and shear force was measured as the indicator of tenderness. Five approaches were employed to characterize the image texture features of pork ham, including the common first-order gray-level statistics (FGLS), run length matrix (RLM), gray-level co-occurrence matrix (GLCM), fractal dimension (FD), and wavelet transform (WT) based method. After that, both simple correlation analysis and partial least squares regression (PLSR) analysis were carried out to study the relationships between the tenderness of pork ham and the extracted image texture features. It was found that the image texture features extracted using the WT-based method had the best relationships with the tenderness of pork ham. However, there were no significant correlations found between the tenderness of pork ham and the image texture features extracted by the traditional methods, including FGLS, RLM, and GLCM (P > 0.05).

Food image segmentation using an improved kernel fuzzy C-means algorithm

In this work, an improved kernel fuzzy c-means (KFCM) algorithm was developed for food image segmentation. Besides the three color components of RGB (red, green, and blue) space, three image texture features were first extracted to represent the spatial contextual information of each image pixel, including polarity, anisotropy, and local contrast. After that, the kernel trick was integrated into the conventional fuzzy c-means (FCM) algorithm to transform the feature vectors to a higher-dimensional space so that the vectors could be linearly separated within this space. Since the original KFCM method was computationally infeasible for real-time implementation, a new scheme was developed to reduce its time and memory complexity by reorganizing the calculations of kernel, membership, and cluster centroid matrices and wiping out the storage of the membership matrix. The experimental results indicate that the algorithm has attractive strengths in generality and computational efficiency.