Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from P-T

Segmentation Evaluation

methods reference standalone using

Definition: Objective segmentation evaluation includes both standalone evaluation methods, which do not make use of reference segmentation, and relative evaluation methods employing ground truth .

The plethora of image and video segmentation techniques in the literature and their wide use in a variety of applications makes increasingly important the need for the objective evaluation of segmentation results. Objective segmentation evaluation can be a valuable tool for supporting the selection of suitable segmentation tools for a given application. It could also serve in joint segmentation/evaluation schemes, to allow for iterative improvement of segmentation accuracy.

Work on objective segmentation evaluation includes both standalone evaluation methods, which do not make use of reference segmentation, and relative evaluation methods employing ground truth. Standalone evaluation methods rely on the calculation of statistical properties of a given segmentation, such as intra-object homogeneity and inter-object disparity with respect to the given features (typically, color information). This alleviates the need for using reference segmentation. In, this fact is exploited in the development of a combined segmentation and evaluation scheme for domain-specific images, utilizing the confidence of region classification results for evaluating segmentation accuracy and forcing the repetition of segmentation using different parameter values until a classification of regions with desired confidence is achieved.

Although standalone evaluation methods can be very useful in such constrained applications, the findings of standalone evaluation methods are likely to deviate from the human perception of the goodness of segmentation, since the latter is largely dependent upon prior knowledge rather that statistical properties alone. Relative evaluation methods overcome this problem by using some form of a reference segmentation result (e.g. a segmentation mask, a binary edge mask etc.) generated by a human observer; evaluation of any segmentation result is performed by comparison with the reference one and calculation of a suitable discrepancy measure. This approach is followed in , where a relative evaluation method using an area-based approach is proposed for the evaluation of still image segmentation results. This method takes into account the accuracy of the region boundary localization as well as under-segmentation and over-segmentation effects, and is shown to be appropriate for comparing segmentation algorithms on the basis of their performance. A related approach to spatio-temporal video segmentation evaluation is developed in, where perceptually weighted evaluation criteria are developed to quantify the spatial and temporal accuracy of segmentation masks.

Segmentation of Images and Video - Spatial segmentation, Temporal segmentation, Spatio-temporal segmentation, Future directions [next] [back] Segmentation and Coding

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or