Appearance Descriptors Research Articles

The genetic basis of human pigmentation

Current methods for forensic DNA analysis require identification of a suspect for comparison purposes. The ability to determine the physical characteristics of the source of a crime scene sample would provide important investigative information. The most obvious descriptors of an individual's appearance are height, colouring and facial features. Since these traits are all highly heritable, as exemplified by the striking similarity of identical twins, it should be possible to determine the polymorphisms that cause the different traits. To date most work in this area has concentrated on pigmentation, since the genetic basis of hair, skin and eye colour is quite well understood. Over 120 genes have been identified in the mouse that affect coat and skin colour and the human homologues of these genes are known. Single nucleotide polymorphisms (SNPs) in a number of these genes have been associated with various human hair, skin and eye colour phenotypes and a number of these SNPs have been shown to have functional affects.

Multilevel Image Coding with Hyperfeatures

Histograms of local appearance descriptors are a popular representation for visual recognition. They are highly discriminant with good resistance to local occlusions and to geometric and photometric variations, but they are not able to exploit spatial co-occurrence statistics over scales larger than the local input patches. We present a multilevel visual representation that remedies this. The starting point is the notion that to detect object parts in images, in practice it often suffices to detect co-occurrences of more local object fragments. This can be formalized by coding image patches against a codebook of known fragments or a more general statistical model and locally histogramming the resulting labels to capture their co-occurrence statistics. Local patch descriptors are converted into somewhat less local histograms over label occurrences. The histograms are themselves local descriptor vectors so the process can be iterated to code ever larger assemblies of object parts and increasingly abstract or `semantic' image properties. We call these higher-level descriptors hyperfeatures. We formulate the hyperfeature model and study its performance under several different image coding methods including k-means based Vector Quantization, Gaussian Mixtures, and combinations of these with Latent Dirichlet Allocation. We find that the resulting high-level features provide improved performance in several object image and texture image classification tasks.

Identifying factors of comfort in using hand tools

To design comfortable hand tools, knowledge about comfort/discomfort in using hand tools is required. We investigated which factors determine comfort/discomfort in using hand tools according to users. Therefore, descriptors of comfort/discomfort in using hand tools were collected from literature and interviews. After that, the relatedness of a selection of the descriptors to comfort in using hand tools was investigated. Six comfort factors could be distinguished (functionality, posture and muscles, irritation and pain of hand and fingers, irritation of hand surface, handle characteristics, aesthetics). These six factors can be classified into three meaningful groups: functionality, physical interaction and appearance. The main conclusions were that (1) the same descriptors were related to comfort and discomfort in using hand tools, (2) descriptors of functionality are most related to comfort in using hand tools followed by descriptors of physical interaction and (3) descriptors of appearance become secondary in comfort in using hand tools.

Sensory Descriptive Analysis of Yerba Mate (Ilex Paraguariensis Saint Hilaire), a South American Beverage

Yerba mate (Ilex paraguariensis Saint Hilaire) is a plant native from South America, consumed traditionally as an infusion called mate. The objectives of this work were to develop sensory descriptors and references for the sensory analysis of the product and generate a sensory profile of 9 Argentine commercial samples from different brands. Eleven descriptors were developed for appearance, 8 for aroma and 11 for flavour. Aroma and flavour descriptors needed specific references for panel training. There were significant differences between the samples for 25 of the 30 descriptors, that could be very helpful for discriminating commercial samples on their sensory properties. Principal component analysis showed high correlation between the following appearance descriptors: 'sediment', 'turbidity', 'brown' and 'dust'. The same stimulus measured both by aroma and flavour, for example 'green aroma' and 'green flavor' were correlated. Samples were differentiated clearly by the first three principal components. (A)

Appearance Properties—A Significant Contribution to Sensory Food Quality?

Word association tests, by (a) presenting food items and collecting free associations from subjects and (b) by presenting descriptors and requesting food products, were performed to establish any relationships between descriptors for appearance properties and food materials. The study demonstrates that appearance descriptors are as frequently associated with food products as texture or flavour descriptors. This underlines the importance of appearance properties such as colour or physical form within the total process of the assessment of food quality. The study also revealed that there are certain descriptors of appearance properties which are strongly linked to specific food materials.

Local appearance space for recognition of navigation landmarks

This paper presents a technique for visual recognition in which the appearance of objects is represented by families of surfaces in a local appearance space. An orthogonal family of local appearance descriptors is obtained by applying principal components analysis to small image windows. These descriptors define the axes of a local appearance space. Each local neighborhood of an image projects to a point in this space. By projecting the set of all neighborhoods of a certain size which compose an image we obtain a discrete sampling of a surface. Projecting neighborhoods from images taken at different viewing positions gives a family of surfaces which represent the possible local appearances from those viewing directions. In this manner we compose a representation in which an object or a landmark can be identified by directly addressing into the local appearance space. Visual landmarks (as well as objects) may be recognized by projecting windows from newly acquired images into the descriptors space and associating them with nearby surfaces. To reduce memory requirements, we propose an efficient tree-search algorithm for the association of points to surfaces. This algorithm directly produces a list of surfaces sorted by distance from the point which represents the observed neighborhood. Our results show that in many common situations, a single window is sufficient to obtain the correct recognition. Furthermore, the confidence of a recognition is easily estimated from the distance from the point to the surface, and the uniqueness determined by the number of surfaces which lie near the point. Robust recognition is easily obtained by reinforcing matching using multiple windows and their mutual spatial coherence. In this paper we present experimental results in the use of this technique for view point invariant object recognition. We then consider the extension of the technique to visual landmark recognition.

Influence of Protein Standardization by Ultrafiltration on the Viscosity, Color, and Sensory Properties of 2 and 3.3% Milks

Permeate and retentate (ca. two times concentration factor) from the ultrafiltration (UF) of skim milk were combined, and cream was added to produce milks with 2% and 3.3% fat and a range of true protein contents of 0.9, 1.8, 2.8, 3.7, and 4.6% within each fat level. A panel that had been trained for descriptive sensory analysis evaluated the appearance, aroma, flavor, and textural attributes of these milks. The relative viscosity and Hunter color values for whiteness (L value), greenness to redness (a value), and blueness to yellowness (b value) were determined.A decrease in the protein content of 2 and 3.3% fat milks by UF made them less white (lower L value), greener (lower a value), slightly less blue (slightly greater b value), and less viscous (lower relative viscosity). A change in protein content had less effect on whiteness (L value) and greenness (a value) of 3.3% fat milk than of 2% fat milk. Within the range of protein percentages studied, no changes in flavor or aroma descriptors were detected for either 3.3 or 2% fat milks. The protein content of 3.3% fat milk could be changed by UF within the range of 1.8 to 3.7% true protein with virtually no change in sensory properties. Protein content of 2% fat milk could be adjusted by UF between 2.8 to 1.8% true protein with no change in sensory scores. A further decrease to 0.9% true protein caused scores for appearance descriptors to decrease. An increase from 2.8 to 3.7% true protein caused scores for both appearance and texture descriptors to increase.

Influence of Protein Standardization by Ultrafiltration on the Viscosity, Color, and Sensory Properties of Skim and 1% Milk

Permeate and retentate (concentration factor ca. 2x) from UF of skim milk were combined, and cream was added to produce skim milk (0.1% fat) and 1% milk with a range of true protein contents (1.0 to 4.8%) within each fat level. A panel that had been trained for descriptive sensory analysis evaluated the appearance, aroma, flavor, and textural attributes of milks. Relative viscosity and Hunter color values for whiteness (L value), greenness to redness (a value), and blueness to yellowness (b value) increased when either protein or fat was increased. The rate of change of L, a, and b values as protein content changed was greater for skim milk than for 1% milk and was greater as protein contents decreased. Sensory scores for several descriptors of appearance, texture, and flavor had a stronger positive correlation with objective measurements of whiteness than with objective measurements of viscosity. Thus, the judgment of panelists about milk texture and flavor in milks that differed in protein content was influenced more by appearance than by viscosity. As the protein content of skim and 1% milk was increased from 2.9 to 4.8% true protein, the sensory properties of the milks were made more like those of higher fat milk, particularly those of skim milk, mainly because of whiter appearance. Panelists perceived changes in the sensory characteristics of both skim and 1% fat milk when the true protein content was increased by 0.9%, the smallest increase that was studied in the experiment.