Jawaharlal Nehru Technological University Hyderabad B.Tech IV COMPUTER VISION Previous Question Paper (December - 2020 - R16)

Texture refers to the visual patterns or spatial arrangements of intensities in an image that exhibit properties of homogeneity. It is characterized by the repetition of basic elements called texels, which may vary in shape, size, color, and orientation. Textures provide crucial information for image segmentation, object recognition, and surface classification.

Key concepts in texture analysis include:

• Statistical Methods: These analyze the spatial distribution of gray levels by calculating local features such as mean, variance, entropy, and contrast.
• Structural Methods: These view texture as a composition of well-defined primitives arranged according to specific placement rules.
• Spectral Methods: These use frequency domain transformations, such as Fourier or Gabor transforms, to identify periodic or directional components within the image.

In the context of texture analysis and spatial filtering, shifts in rectangular neighbors refer to the relative displacement vectors used to compute spatial dependencies between pixels. These shifts define the spatial relationship between a reference pixel and its neighbors within a rectangular grid, often used in the construction of Gray-Level Co-occurrence Matrices (GLCM).

The shifts are typically represented as a displacement vector (Δx, Δy). The common shifts for rectangular neighbors are:

• Horizontal (0°): Defined by the shift (1, 0), where the neighbor is located immediately to the right of the reference pixel.
• Vertical (90°): Defined by the shift (0, 1), where the neighbor is located immediately below the reference pixel.
• Diagonal (45°): Defined by the shift (1, 1), where the neighbor is located at the bottom-right corner.
• Anti-diagonal (135°): Defined by the shift (-1, 1), where the neighbor is located at the bottom-left corner.

These shifts capture directional texture properties. By varying the magnitude of these vectors, one can analyze textures at different scales, effectively capturing both fine-grained and coarse-grained spatial correlations.